April 2, 2025 • 46 mins
What exactly is cloud networking? This seemingly simple question quickly descends into a fascinating philosophical debate as we welcome back Nico Vibert, Senior Staff Technical Marketing Engineer at Isovalent/Cisco, to tackle this identity crisis head-on.
The conversation begins with a startling observation from Nico about analyst reports that group wildly different vendors together under the "cloud networking" umbrella. From there, we explore how defining cloud networking has become increasingly complex as technologies evolve and converge. We trace the origins back to AWS's introduction of VPC in 2009 and discuss how different cloud providers approach networking based on their unique company cultures.
One clear consensus emerges: true cloud networking must be API-driven. Whether consumed directly via APIs or through infrastructure-as-code tools like Terraform, programmability stands as a non-negotiable requirement. But beyond this foundation, the boundaries blur significantly when examining various technologies that might qualify.
Does Kubernetes networking fall under the cloud networking umbrella? What about Middle Mile providers like Equinix or Megaport that physically connect clouds? Are CDNs part of cloud networking, or something entirely different? We dissect these questions without settling on definitive answers, highlighting how technology's rapid evolution makes categorization increasingly difficult.
Looking ahead, we explore how AI is reshaping cloud networking in two critical ways: networks optimized for AI workloads and AI-enhanced network management. Cloud providers are investing billions in infrastructure upgrades, developing custom silicon to reduce dependency on GPU manufacturers, signaling massive transformation on the horizon.
Whether you're a network engineer, cloud architect, or technology leader trying to understand this evolving landscape, this episode provides valuable perspective on cloud networking's past, present, and future directions.
Connect with Nico: https://www.linkedin.com/in/nicolasvibert/
Purchase Chris and Tim's new book on AWS Cloud Networking: https://www.amazon.com/Certified-Advanced-Networking-Certification-certification/dp/1835080839/
Check out the Fortnightly Cloud Networking News
https://docs.google.com/document/d/1fkBWCGwXDUX9OfZ9_MvSVup8tJJzJeqrauaE6VPT2b0/
Visit our website and subscribe: https://www.cables2clouds.com/
Follow us on BlueSky: https://bsky.app/profile/cables2clouds.com
Follow us on YouTube: https://www.youtube.com/@cables2clouds/
Follow us on TikTok: https://www.tiktok.com/@cables2clouds
Merch Store: https://store.cables2clouds.com/
Join the Discord Study group: https://artofneteng.com/iaatj
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Tim McConnaughy (00:14):
Hello and welcome to another episode of
the Cables to Clouds podcast.
I'm your host this week, tim,and with me, as always, is my
co-host, chris.
I'm at Carpe DMVPN on Blue Skyand Chris is at BGP Main on Blue
Sky, if I could speak correctly.
And with us is a returningguest, a good friend of the
(00:36):
podcast.
We're lucky to have him.
Nico Weber, would you like to?
Actually, I don't remember,nico, are you on?
I don't remember.
Are you on Blue Sky yet, nico?
Are you?
You to actually?
Nicolas Vibert (00:44):
I don't remember .
Nico, are you on Blue Sky yet?
Nico are you?
You don't do the social media.
No Well, I used to be onTwitter and pretty active, but I
left that, yeah.
Tim McConnaughy (00:52):
That was a good move.
Nicolas Vibert (00:55):
Pretty active on LinkedIn with videos and silly
blog posts, but that's whereyou'll find me.
Tim McConnaughy (01:01):
Yep, yep, yep, excellent.
And just in case, so we've hadyou on the podcast before.
It was an excellent episodetalking about Kubernetes and
Cilium, but for the people whomay not have heard or seen that
episode, go ahead and quicklyreintroduce yourself.
Nicolas Vibert (01:17):
Sure.
So I'm a senior staff technicalmarketing engineer at
iSurveillance slash Cisco and mybackground is network
engineering.
I've been in network for about20 years now.
I did my CCNA, ccnp, ccie backin the days.
I was CCIE for about 10 yearsuntil I let it go and worked for
(01:39):
Cisco for a while and then Iwas a boomerang back in through
the acquisition of iSurveillanceby Cisco last year.
So yeah, just happy to be backon this podcast.
Tim McConnaughy (01:52):
Yeah, we're definitely happy to have you
back and actually you came upwith a really cool topic and
that's why we brought you back.
And let me see if I can readthis correctly, just make sure
I'm getting it right.
So we were talking on LinkedIna little bit and you were like
you know, what I really love totackle was this idea of the
identity crisis of cloudnetworking, specifically, what's
cloud networking and what it is, what it isn't.
(02:14):
And I just realized that, likeyou know, there's so many
opinions on this that, like it'sa great topic and I'm sure if
you talk to 10, five engineers,you'll get 10 different opinions
, right, so I love it.
So let's just kick it off there, like what made you think of
this topic?
And like because I mean, sothis comes from somewhere.
You've got opinions on this,yeah.
Nicolas Vibert (02:41):
Well, I guess, as a marketing, technical,
marketing engineer, one of myresponsibilities is to work with
analysts, you know, like thelikes of Gardner and Forrester,
that kind of people, and answerquestionnaires about our product
and where we fit, and obviouslylet them, you know, rank
stackers against other companies, you know, in the quadrant
right.
And I was recently answeringone such query about cloud
(03:04):
networking.
I looked at all the differentvendors and I was like, hang on,
are these guys cloud networking?
And you just look at the listof people we were being compared
with, it was like apples andoranges, lemons, I don't know.
It was just a bunch ofdifferent companies and I just
kept thinking what is even cloudnetworking?
And you guys, you've had likecloud networking and you guys,
(03:24):
like you know, you've had acloud networking podcast for a
couple of years now a coupleyears, yep yeah, um, you
probably, even between yourself,have a different definition of
what cloud networking might be,so no, it's.
Tim McConnaughy (03:38):
It's funny to hear you say this because since
I became a TME, one of thethings I probably hate most are
those 40-page reportquestionnaires about hey, here's
, fill out 40 pages of why youfit into this quadrant and how
you're better, and you know,usually they're like, oh, don't
talk about your competitors inhere.
But then you get to see, likeyou said, like you go look at
(03:59):
the magic, the quote, unquotemagic quadrant or whatever.
They've all got their ownversion of stack ranking vendors
and you're looking at them likehuh, Like why am I?
These guys are like SSE orsomething.
They got nothing to do withwhat I do.
Right, so absolutely All right.
Nicolas Vibert (04:16):
So I'll ask a question then who do you think
was on this cloud networking,you know?
Who else we have ISOVANON,which, as I said, is now part of
Cisco.
There might be some that youthink, oh yeah, of course that's
a cloud networking company, andthere might be some that was
like, hang on, so give me your,give me your, give me your
(04:36):
guesses.
Chris Miles (04:37):
Like versus like versus, who is versus who isn't.
Nicolas Vibert (04:39):
Well, who do you think is on it?
Tim McConnaughy (04:41):
There's a list.
Nicolas Vibert (04:45):
Who would be on the list?
Yeah, I've got the list.
I'll tell you if you're.
Chris Miles (04:47):
If you're correct, or not nice uh yeah, well, I
mean, I would hope.
I would hope our currentemployer is on there as well.
Tim McConnaughy (04:53):
Uh, I'm sure a matrix has got to be on there
somewhere.
Yes, uh, I mean so obviously,alkira would I.
I mean I say obviously, but I Ibelieve that alkira would would
qualify and I feel like we'relike passing judgment on that.
I feel like an asshole, we'rejust passing judgment on these
companies.
But no, I mean I think from afocused cloud networking
perspective.
Yeah, I mean I think Alkirawould fit the bill for that.
(05:14):
Prosimo probably did before itwent under right.
I mean obviously Isovalent.
Did Calico's own?
Is it Tigera's the company thatactually does Calico?
Is that right?
Nicolas Vibert (05:30):
Yeah, they're actually not on it, but I
suppose they could be, becausethey offer a similar product to
what we do.
Tim McConnaughy (05:35):
Yeah, well, okay.
Now, this is, of course,because I only just started
getting into the Kubernetesstuff within the last few weeks
and I went straight to Ciliumbecause I already knew you and I
was very familiar with the labsand everything that you pointed
me towards.
But then I was readingsomething else and I was like,
oh, I should take a look atCalico and it's not the same,
and I don't want to compare thetwo products.
(05:55):
But I was like, just from a,you know, it's a CNI plugin,
right, like it fits at least theKubernetes CN, kubernetes, cni,
I that's.
And again, this gets into thewhole cloud networking thing.
Well, which you know, it'skubernetes, kubernetes
networking, which could berunning in the cloud.
So, yeah, you can see how itautomatically.
Nicolas Vibert (06:10):
Already we're starting to draw lines, right,
right yeah, and you know, like Ithink if I, if I had to guess,
I would have said aviatrix andprosimo, now kira, because
they're kind of for me likeobvious networking companies, um
, but then and then I saw, okay,I saw cisco and arista, juniper
(06:31):
.
Fair enough makes sense.
Tim McConnaughy (06:33):
So f5 yeah, that's yeah, are we?
That's like a fair point, right?
They route packets but their,their actual value, prop and
application is in the loadbalancing, right.
So that's a fair point.
Uh, oh no, didn't f5 boughtsomebody though?
Uh, a few years ago, um, whatwas the name of that company?
They bought a they bought acloud.
Chris Miles (06:53):
Yeah, was it engine x.
Nicolas Vibert (06:54):
Sorry, it's a bit of gen x yeah yeah, I'll
have to look it up.
Tim McConnaughy (06:58):
They bought an actual company that they were
like, but then they kind ofburied it.
They never came out with theproduct, the the.
They actually had aquote-unquote multi-cloud
product and they never came outwith it.
Nicolas Vibert (07:07):
Now.
I have to go find it Well.
Tim McConnaughy (07:08):
Cloudflare actually just today had a press
release about a new MCNS product.
Did you see it?
Oh, did I?
Okay.
Chris Miles (07:16):
Yeah, they acquired a company called Nefeli I think
it was about a year ago, rightand now it's being called ago,
right, um, and now it's beingcalled like of course they've
dubbed their, their brandedmagic, uh, moniker onto the
front of it now, but it justlike that seems like an
orchestration thing.
So but like I think, as we openthis up, that kind of speaks to
your point, nico, like what iscloud networking?
(07:38):
Like?
We have like companies justdoing maybe like kubernetes
networking and then maybemulti-cloud networking, and then
you know something as simple asjust orchestrating the native
constructs, like is that a cloudnetworking company?
Like it's uh, yeah, the, themoney.
Nicolas Vibert (07:53):
the waters are very muddy at this point yeah,
you know, I see, like the likesof you know broadcom vmware is
that.
You know, do they includevmware and I make vmware and
yeah, you know, like an sx orsomething.
Yeah, I'm assuming that's an sxright I would think so yeah,
but like would you?
would you describe an sx ascloud networking?
(08:16):
And you know, and again, Iworked on an sx and which is I?
I mean an sx couldn't be moredifferent than you know, like
what, what, what I don't know,f5 or prosimo, or if hvc says
right.
So this is what I mean about,like this, cloud networking
products and companies thatactually do very different
things.
Chris Miles (08:35):
Yeah, yeah, I feel like.
I feel like this is almostsomewhat adjacent to like the
kind of the evolution of wheresd went.
Oh yeah, we had a few companiesstart out that were obviously
pure SD-WAN companies, like youknow, kind of in that kind of
founding sector, so to say.
Tim McConnaughy (08:52):
Like Viptala Of the product.
Chris Miles (08:54):
Versa, versa, yeah, and there's one other one
Avello Cloud.
Tim McConnaughy (09:00):
Yeah, Avello.
Chris Miles (09:01):
And then, but you know, then we started to see the
likes of the major networkingvendors.
Uh, just start like maybe ormaybe, like I saw, mainly
networking vendors, and like wayin optimization they were just
yeah the way in ops too, yeah,firewall vendors, firewall
vendors, they would add like onefeature and now it's like okay,
well, now this is sd-wan it'slike I don't know I don't know
(09:22):
if you can really you, you knowput your merit on that Like
that's just like.
Nicolas Vibert (09:26):
Yeah, and it was like it went from like being
SD1 to being SASE to being ZTNA.
Tim McConnaughy (09:33):
you know it's so quickly.
Nicolas Vibert (09:36):
But we were like yeah, like sorry, it was like
even SD1, like, is that cloudnetworking?
Because you know, you used tothink about SD-WAN and you had,
like, you are your branch deviceand you would say, oh, let's go
and take all that traffic toOffice 365.
And instead of backhauling backto my data center and then
through my internet pipe, I sendit straight to my, you know,
(09:56):
over my local internet link.
That's some form of kind ofcloud networking as well.
Tim McConnaughy (10:01):
Not, you know, not the yeah, well, this is
interesting too, because the uhthis is where we get into some
other things where there's kindof a long-standing curiosity
about like is saz cloud the samelike?
You know, like when, uh, wetalk about this all the time,
where, like customers are like,or microsoft is like, we have,
uh, you know, so many millioncloud customers and it's like,
(10:21):
well, I mean they're, they'reoffice 365 customers.
I don't know if I would callthem cloud customers.
Nicolas Vibert (10:24):
And it's like well, I mean they're Office 365
customers.
I don't know if I would callthem cloud customers.
Tim McConnaughy (10:27):
Sure, it runs on Azure, but I don't know if
that makes them Azure customers.
You know what I mean.
Nicolas Vibert (10:32):
Like it's not quite the same, Let me ask you
guys, like where do you thinkcloud networking was invented?
Tim McConnaughy (10:41):
I mean like the CSPs.
Chris Miles (10:43):
Yeah, I mean mean ultimately, I think the.
If we look at the evolution ofcloud, you know kind of start in
the early days with, you know,launch of S3, and then you know,
eventually more things startedor like using cloud and needed
the natural progression andeventually things needed to be,
you know, able to talk to eachother, right, I don't remember
what was it.
When did VPC launch?
Was like 2010.
Wow, literally.
So that's to be you know, ableto talk to each other, right and
(11:04):
um, I don't remember what wasit.
When did vpc launch?
Was like 2010, 2000.
Well, literally.
Nicolas Vibert (11:08):
So that's to 25th of august 2009.
I found jeff biles uh,announced some blog posts.
Oh, nice, nice, um, and it'sjust, you know, you.
You read it and it's like, wow,you know.
That's like close to 16 yearsago and all still relevant,
well-written, clear.
The blog post was about, youknow, being able to create a
(11:31):
virtual isolated networkconnecting back to on-prem with
a VPN, so that for me that feelslike the origins and I'm sure
some people will disagree, butit feels like the origins of,
like, cloud networking.
Tim McConnaughy (11:45):
Well, here's the question, though Hold on
real quick, because we got to.
Some people will disagree, butit feels like the origins of
like cloud networking.
Well, here's the question,though Hold on real quick, cause
we gotta.
Nicolas Vibert (11:53):
I'm waiting for the data center people to start
screaming that like privatecloud.
What about private cloud?
Yeah, and you think okay.
Tim McConnaughy (12:03):
I think the response to this wasn't, you
know, things like OpenStack.
Yeah, that's a very good point.
Actually, I like OpenStackright.
Yeah, that's a very good point.
Actually, I think OpenStackwould probably be, because, I
mean, we're really talking abouta model, we're not just talking
about private cloud as a datacenter.
That's not you know its ownkind of abstracted workflow.
Because to get to that onprivate, yeah, I think the first
thing you would have been ableto work with to do that would
have been, yeah, openstack, totry to abstract away the compute
(12:26):
and stuff.
Nicolas Vibert (12:27):
And they had the Neutron or Quantum plugin which
was their way of creatingvirtual network and again I
think that's my recollectionwhere it was in response to the
cloud providers and taking offand they wanted to cloudify the
way networking was beingconsumed, on-prem right To offer
a similar kind of experience.
(12:48):
Yeah, it's again.
We still haven't really got tothe definition of cloud
networking, right?
Tim McConnaughy (12:54):
No, no, no, no, we'll get there.
We'll get there around 55minutes or 60.
Chris Miles (13:01):
Well, it's funny because I think cloud has always
been used pretty much as anabstraction.
Right At the end of the day,there's still kind of, you know,
hardware and components runningunder the hood that we all are
familiar with.
But I feel like, yeah, thecloud initially was supposed to
be this term.
That was like somewhat agnosticand you know, we had private
(13:21):
cloud, we have public cloud, but, like, private cloud is a very,
very different experience thanthan public cloud and I feel,
like the terminology has kind ofjust shifted cloud I mean.
In my mind cloud means publiccloud, um, and what that
ultimately translates to is, youknow, basically using, uh,
vendor or service providerspecific APIs to interact with
(13:44):
their environment to get it todo what you want to do.
So it's like I mean, that'swhat it sounds like to me.
I mean, ultimately, you'realways going to be able to do
what they give you access toright.
Yeah, just like MSP, right,there's no super level of
customization that you could dowith a private cloud Right.
So I mean, that's the way ittranslates to me.
Nicolas Vibert (14:05):
And I think that for me, cloud networking, it
does involve some form of publiccloud, because otherwise the
definition just becomes toobroad.
Yeah, it's too generic.
Chris Miles (14:18):
Yeah, Otherwise it's just networking.
Really, yeah, networking.
Nicolas Vibert (14:25):
But I think what you said around consuming the
MSPs APIs, I feel like the firstkind of real form of kind of
cloud networking companies, likethe likes of Aviatrix, was
really about like building a gapthat the cloud networking, the
MSPs where the networkingconstructs had some limitations.
(14:48):
So the startup was like, oh,let's go and create a product
that fits this gap.
That's fair.
Tim McConnaughy (14:54):
Yeah, no, I couldn't agree more, 100%, 100%
yeah, it's.
In fact, if you look at thehistory of our company, if you
look at the features we rolledout, first stopgap, fill in the
blank.
Like these don't exist in thehyperscaler and I think there
will always be a limitation towhat the hyperscalers can offer
to customers.
Just by the fact that they'rehyperscalers, right, they can't
(15:14):
give you the same granular levelof control that you can give to
if you're going to, if theygive it to you, they got to get
to everybody, right.
Chris Miles (15:20):
So that's yeah, because it's a service, right,
like, like hardware, likehardware vendor can give you
something off the shelf thatdoes everything under the hood,
because you control all of thatand you know it's probably in
your isolated environment, right.
But if you're providing aservice, there's only so much
you can do, um, for, uh, youknow, like with r&d development,
things like that, it just takesso much more work to get to
(15:40):
that, get to that level I thinkevery uh, even every msb, has a
different stance and culturetowards networking.
Nicolas Vibert (15:49):
I mean, I feel like AWS was, like you know, so
developer-centric and I don'tknow, google is, like, so proud
of its networking, so proud ofits backbone and, you know,
happy to share.
Oh look, we've got cables, youknow, under the Pacific Ocean,
the pacific oceans.
Look at that, it's amazing.
And like they're just very,very proud of someone's
networking and their pipes andtheir innovation around this.
(16:12):
Uh, they take now, that'sinteresting yeah, they have a
different perspective, I think.
I think that's a culturalaspect to uh, to to the
importance of networking yeahthe like.
Chris Miles (16:23):
it's always cool when they show those things, but
in my mind, whenever I see youknow, um, you know a service
provider, specifically a cloudservice provider, kind of toting
about like what kind ofconnectivity they've built with
you know undersea cables, allthis kind of stuff, all that
says to me at the end of the dayis like, there's capacity, like
the like.
Capacity will probably never bemy problem, which is good.
(16:45):
It doesn't change how I buildnetworks and I'll be honest with
you in the long run, the leastinteresting thing well, not the
least, but one of the most leastinteresting things to me in
networking has been QoS, and ifI don't need that anymore, then
I'm very happy with that.
Nicolas Vibert (17:01):
So I think this is progression for sure.
Chris Miles (17:05):
But I don't know.
Like yeah, networking to melike the undersea cables that's
all cool, the CDNs that's coolas well, but like I'll be honest
like that doesn't move theneedle for me, like in terms of
cool technology.
Tim McConnaughy (17:18):
So it's interesting, nico, you have the
observation that Google cares alot about its networking,
because I think that's true.
What you said about is buildingthe undersea cables.
I think all that is true, but,like, when I look at google's,
the way that you as a customerhave to consume google's
networking, to me it seems likeextremely developer centric.
Like like don't even theconcept of global vpc and
(17:41):
specifically the idea that likewe can just do it all for you.
It seems like they really justwant you to not care about
networking, or just for meworking with it.
It seems, you know, like theauto building uh vpcs where
they'll auto do all your subnetsfor you.
It seemed to me like they theywere focused on more.
So it's interesting that theopinion that you have that is
different yeah, it's.
Nicolas Vibert (18:02):
I mean to be fair, like I um, I was at cisco
last week and I was showing ademo with a couple of colleagues
that uh, which was like ummeshing a couple of capabilities
cluster one was on-prem, onewas um in, uh, the cloud and we
picked, we picked uh, gk, uh,because we thought actually from
(18:23):
a networking perspective, thatwas the easiest.
Tim McConnaughy (18:25):
Oh, for sure, I think so.
With the global VPC andeverything and the way they do
it.
I think GKE is the easiest.
Nicolas Vibert (18:32):
And you know, doing some cool connecting
machine clusters together, doingsome low balancing across sites
.
That's a nice form of cloudnetworking.
Tim McConnaughy (18:41):
Yeah.
Nicolas Vibert (18:42):
It's interesting .
Tim McConnaughy (18:42):
I'm actually just getting into this myself
because we're kind of doing somestuff with Kubernetes as well,
so I've been so late, Like Iknow we had you on what like
almost a year ago now orsomething like that.
It's been a while and I was oneof those things where I was
like I'll get around to it.
I didn't really get a chanceuntil a few weeks ago, but the
more I learned about it, themore interesting it is,
especially the networking aspectof it about it, the more
(19:05):
interesting it is, especiallythe networking aspect of it,
even though it's like 90% of itis abstracted away, at least
within the cluster, certainlywithin the cluster it is.
Nicolas Vibert (19:11):
Which is interesting from my perspective.
It was my first time presentingat Cisco Live.
I've been to Cisco Live manytimes but I've been presenting
at KubeCon, for example yeah,and presenting our networking
platform for Kubernetes.
It's very different when you goto somewhere like Cisco Live.
(19:32):
Compared to KubeCon, audienceis very different.
Yeah, exactly, very, exactly,very different audience and,
like some of the KubeCon,they're more interested in like,
what can you do with all thisnetworking?
Like you know more about theapplication stuff, can you do
things like canary testing andload balancing and introducing
new version of my app, do someredirection?
(19:54):
Lots of different, more, moreagain application-centric,
whereas I go to Cisco Live andwhat the first question I get is
like, show me your packet work,show me the headers, and you
know, show me BGP, you know, andit's just like oh boy, but so,
yeah, it's kind of again adifferent culture.
But what was?
(20:14):
I guess?
What I found fascinating is theplatform and Kubernetes.
Networking is aboutcloud-native applications, which
is mainly containers and mainlymanaged and orchestrated by
Kubernetes, and I think that'sone of the reasons we were
invited in this vendor's analystreport is because we do
(20:35):
networking for cloud-n nativeapplications.
So does it qualify?
As you know, is that arequirement to be cloud
networking?
Do you need to be able tosupport modern containerized
applications, or is that just akind of a side aspect of it?
That's a great question.
Chris Miles (20:53):
Yeah, I mean, I think like we're at this point
where we probably like I feellike we don't want to fall like
in the same direction, likemaybe security has where there's
, you know, there's so manydifferent categories for things,
like there's, you know like somany acronyms, cnap, all this
kind of stuff.
There's probably been twoannounced today as yeah, there's
(21:15):
a lot um but like I don't knowif we wanted like venture into
that territory where we havelike ultimate levels of
classifications to every one ofthese things, like um, I mean,
we're already there, we cangenerally talk about it, but
there's just no, you know, likeofficial, you know, once gartner
does it, then then then it'sreal right um, but yeah, I don't
know if I wanted to go that,that route, but I mean,
ultimately there there isconfusion.
(21:36):
So you know, the consumer at theend of the day, is going to be
the one that's confused whenthey hear this term.
Right, so, um, it might benecessary, I don't know.
Tim McConnaughy (21:44):
So, so this, uh , you know you get me thinking,
though, because, like Kubernetes, the whole thing with
Kubernetes is that we, you knowyou can have this cloud native,
Cloud native and Kubernetes likego together.
Right, the CNCF owns the wholething.
Like there's so much it'sjoined at the hip.
Right, we say cloud native,cloud native, cloud native.
You kind of wondered, you know.
Now we have to like, okay, whatthe hell does cloud native mean
(22:06):
?
In order to get to cloud native, what the hell is cloud native?
We had a discussion on this aswell we did, we did a while back
.
Chris Miles (22:13):
Yeah, it's like I think it's strayed from what it
what it used to mean or what itwas intended to mean as you can
force.
Tim McConnaughy (22:19):
You can run a kubernetes cluster on prem right
or, you know, in a middle mileprovider or something.
I don't think I'm not, I don'tknow how many people do that.
To be honest with you, I thinka lot of people consume the, the
managed kubernetes.
Kubernetes is prettycomplicated, but well, actually
I mean the.
Nicolas Vibert (22:33):
I mean just again talking to customers at
Cisco Live and KubeCon.
So many are doing self-manageand you know, very often on the
like, you know like OpenShift orRancher, and that's why I had
so many conversations aroundthis last week at Cisco Live.
Again, it's about like okay,I'm building my own Kubernetes
(22:54):
cluster, managing my own on-prem, how do I connect to my Cisco
ACI, for example?
Yeah, of course, any otherfabric.
Tim McConnaughy (23:06):
Just outside the cluster.
How are we making connectivityright?
Yeah, how do we expose our?
Nicolas Vibert (23:10):
applications applications.
What's the best practices?
Again around BGP and Ingressand make our cloud native
applications accessible from theoutside.
Tim McConnaughy (23:25):
Yeah, it's funny because with Kubernetes,
specifically for Ingressspecifically, you have to build,
you need a load balancerservice, something for the
ingress to the service you'vedone.
So now a load balancer nowbecomes part of cloud networking
, Like all these services.
These extra services end upbeing part of cloud networking.
(23:46):
Right, it's not just networkinganymore, there's all these
services included.
Nicolas Vibert (23:51):
Yeah, so is service mesh a service mesh or
an ingress controller?
Is that cloud networking?
Tim McConnaughy (23:59):
I think it has to be with Kubernetes
specifically?
I think it has to be right, theservice mesh especially.
You've got your sidecar.
You're building tunnels.
You're just making a tunnelmesh between clusters with your
sidecars.
Chris Miles (24:12):
That one almost ventures just into application
networking.
Nicolas Vibert (24:15):
to me, yeah, maybe At the end of the day, I
guess like what is that that's?
Chris Miles (24:20):
fair.
I mean it's turtles all the waydown right.
Eventually it's the same thing,but like I don't know, we're
probably splitting hairs.
Tim McConnaughy (24:28):
No, I mean, that's a fair point though
Application, because it's theapplication.
So yeah, I mean now that you'reasking yourself, god, this is
getting so crazy.
Nicolas Vibert (24:36):
So, philosophical.
Tim McConnaughy (24:38):
Yeah it's almost Socratic in the method,
right?
Because then you have to saywell, is the Envoy like for
Istio, is the Envoy anapplication or is it a
networking?
Nicolas Vibert (24:51):
device or is it both?
Yeah, yeah, yeah, I mean it'sproxy-wise.
So yeah, it's kind of.
You know, defining a categoryis just really hard when it just
keeps things keep evolving.
But I think for me, I guess,the characteristics of cloud
networking.
There are things that I thinkwe can say, okay, the one thing
that all have in common for meis that they're API-centric.
(25:13):
That would be, I guess, my ownpersonal take, and maybe you
know I'm sure there are alsocloud networking products that
are maybe more consumed througha UI, but I think the aspiration
should be that you consume yourcloud networking.
Your networks are API, builtthrough an API.
Tim McConnaughy (25:34):
Okay, okay, yeah, I think that's Well.
What about Terraform?
Did you consider that the samething?
It's IOC, right?
Nicolas Vibert (25:41):
Yeah, with an API, some form of automation, a
playbook, a Terraform module,okay, yeah, yeah, that should be
part of the goal when you startleveraging a cloud networking.
Part of the goal when you startleveraging a cloud networking
is that you template, create atemplate or automate it.
If you claim you have a cloudnetworking product and you don't
(26:07):
have an API, then you're lying.
Tim McConnaughy (26:12):
Or, at the very least, if your product is not
interacting with the cloudprovider APIs to do the building
, like that kind of stuff.
Yeah, I think that's definitelytrue.
Yeah, I think after I askedabout the Terraform question, I
realized that what youoriginally had meant when you
said that so yeah, it is APIsall the way down, right, that's
how you generally interact with,whether you're doing it on a
(26:34):
CLI, whether you're doing it ona CLI, whether you're doing it
with their SDK, whatever.
That is right.
Terraform, it's an API call atthe end of the day.
So you asked a question on thenotes here that I really wanted
to get to and I don't want topick out a specific company, but
, like you had said, essentiallyis Middle Mile a cloud
networking company, somebodythat provides connectivity to
(26:56):
the cloud?
Would we consider them a cloudnetworking company.
Nicolas Vibert (27:03):
You have an opinion, yeah, you could argue,
and again you could say thingslike Equinix, megapod.
They provide some direct linksto the cloud from on-prem to the
cloud.
I think they also enable you toconnect your clouds together,
so some form of inter-cloudconnectivity.
(27:27):
Now, I think because it's moreof a physical layer to and from
the cloud.
I don't think it necessarilyqualifies as such.
I guess in my head it's more.
I guess cloud network, in myhead, is more about creating
networks within a cloudenvironment that isn't consumed
(27:52):
by applications that live in thecloud and that can be virtual
machines, that can be bandwidthor that can be containers.
Chris Miles (27:59):
Okay, so real quick then I was going to say an
important thing in thatclassification is they also
provide the connectivity likedirect connectivity options for
like SaaS providers as well.
Right, so that's like servicesrunning on cloud, but maybe
you're not getting into aprivate environment that you
manage on top of cloud.
So it's like running on cloud,but maybe you're not, you know,
getting into a privateenvironment that you manage on
top of cloud.
So it's like, yeah, like whatdoes that translate to?
Tim McConnaughy (28:23):
Well, I also have to challenge a little bit
the idea that if Megaportconnects two clouds together but
it's physical, that makes themnot cloud.
Because that then question isso if I connect AWS to Azure and
I do it over the internetinstead of using, because I mean
, you know, there's no directconnect between AWS and Azure.
That's obviously right.
(28:43):
So is that cloud networking orit's going over the internet?
Are we counting the internet asa medium?
You know what I mean.
Chris Miles (28:51):
But if we're thinking about like this concept
of at least the CSPs providednetworking options, you're
basically building networkingacross their backbones.
Whatever services they'veexposed, they allow you to build
networking on their platform.
What is different about that toMegaport or to Equinix?
That's the exact same thing.
(29:12):
They have an underlying set ofhardware services, a backbone of
something.
But the way you interact withit is essentially the same.
You make an API call and yousay build this connectivity.
So how is it not cloudnetworking at that point?
Tim McConnaughy (29:26):
We're not saying you're wrong.
I'd like to justify a littlebit.
Nicolas Vibert (29:31):
I suppose it's.
You don't necessarily use theirnetworks to run applications.
Tim McConnaughy (29:39):
Okay, so it's that you're not building apps on
their network?
Nicolas Vibert (29:42):
Yeah, and maybe I don't know enough about what
they do.
I guess it almost feels like asubcategory of cloud networking.
Maybe this is how we kind ofbreak it down right yeah clam
cloud networking as a kind ofoverall category, and then
you've got the intercloud, andthen you have the I mean it's
(30:05):
just not easy, right likethere's a reason why everybody
struggles with it.
Tim McConnaughy (30:08):
This definition , these definitions, is not easy
.
I think it's very hard to todraw these lines and to figure
out what lines to draw.
Nicolas Vibert (30:15):
And you know we talked.
You were saying like Cloudflareor maybe like CDN I don't
necessarily Akamai.
Tim McConnaughy (30:22):
Yeah, that's a really good question, man.
That's a super good.
That's even harder than theMegaport one, by far, I think,
or Megaport Equinix Middle Milegenerally by far, because, you
know, the CDNs are technicallyusually attached to the cloud
backbone, but like they're not,so they're, they're delivering
content right.
That is, over a network that isusually cloud based in some way
(30:46):
, whether it be Cloudflare ordirectly connected like with the
AWS offers.
Man, I don't know what do youthink?
Chris?
That's a tough one.
Chris Miles (30:54):
Yeah, it is tough, but I mean that one one, that's
one that I feel like has beenaround in some shape or form for
a long time as well, like evenbefore cloud oh yeah, cdns, yeah
yeah well, like I mean, yeah,like I remember, like back in
the day we used to, I used towork at a big mpls provider and
like we got a lot of circuitsfrom level three and at level
three, had a lot of these kindof front front door scrubbing
(31:17):
services and things that arejust very adjacent to what those
providers do today andobviously they have a much
bigger footprint and things likethat.
So it's like I don't know Thenso does it?
If that's the case, then thecloud networking start back then
, like yeah yeah it is, it is,it is, it is, it is, it is.
Nicolas Vibert (31:35):
Yeah, yeah, it is, it is.
It is very hard to say like,yeah, what, what was?
What started it all?
Um, yeah, okay, so what aboutsecurity?
Then?
Like, do we is, let's say,consistent security across cloud
?
Is that a must?
Is that like um, because wecan't do talk, we can't talk
about cloud networking withouttalking about cloud networking.
(31:57):
Security Is encryption, isconsistent networking across the
multiple clouds?
Yeah, some form of requirementto say, okay, we do cloud
networking.
Tim McConnaughy (32:12):
So, yeah, obviously, we work for Aviatrix
and we literally this is kind ofthe mission statement of the
company.
So, to put the company aside, Ithink, even putting the company
aside, I think you know ifyou're gonna this whole shared
responsibility model that allproviders have, you know,
customers are still in charge oftheir own data, right, the data
(32:35):
that's runs across their cloudnetworks.
Now it's encrypted in the rack,essentially, or across the
backbone, or however you want tosay it on the cloud provider
side, but ultimately a customer,but that you know.
Then you look at things likesalt typhoon uh, the sulfide
typhoon attacks, uh, you knowowning telecom networks and
being able to just do man in themiddle, you know captures of
(32:56):
open, of unencrypted traffic.
So, yeah, I mean, I do think Ithink it's a requirement, in
that a customer would be veryfoolish not to run some of their
, you know, do it themselves.
I wouldn't trust the providerto secure all of my traffic, but
does it?
Is it quote unquote required?
Like from a technical RFCperspective, if you will?
(33:18):
I mean, obviously not right.
Like AWS supports GRE acrossyou build a GRE tunnel from a
TGW Connect, for example, and ofcourse you can't do GRE in
Azure, but that's because that'swhat they're using.
On the underlay right.
Chris Miles (33:32):
It's funny how many directions this can go in,
because that's the first thoughtyou had, tim, about this.
I call back to just when we'retalking about the different
architectures that we see withincloud and how it's a very
strong fundamental shift withwhat we've been doing for 20, 30
years at this point.
Tim McConnaughy (33:52):
Oh yeah, the Volus coupling model.
Chris Miles (33:53):
Yeah, well, not even that.
I'm just talking.
If we think of the concept of aglobal VPC, like that's one big
network that exists everywhere.
Like the difficulty toimplement something in there,
like zero trust, which is aprominent thing you know from
from a networking perspective.
Like doing that in a globalconstruct that is just all
meshed together at one point,like that becomes a conversation
(34:15):
.
That's very difficult to getthat in there, right?
Tim McConnaughy (34:19):
What's your opinion, Nico?
Nicolas Vibert (34:22):
I can't imagine doing like a global VPC back
when I used to be like a networkoperator.
I just imagine there's adifficulty of doing something
that is so well crafted that youknow like you you get, you know
you deploy VPC in GK and youdon't really care Like you, just
(34:46):
you know your applicationreceiver, it just works and you
know.
And that's just thinking again,thinking back to the days where
I had to connect multiple,multiple sites together and
having to worry about addressing.
And I think that's the beauty oflike a really well designed
cloud networking like GKE, butagain also, providers also do it
(35:07):
very well, but it's all theintelligence and the automation
under the hood to, you know,make it happen and being able to
replicate it on-prem is justabout impossible.
Tim McConnaughy (35:20):
We've said that for even when Alex was here,
that we used to talk about allthe time that the people that
are going to win the game aregoing to be yeah, I know who is
that?
I don't know who that is eitherthe people who can, whatever
company or companies finally areable to build the product that
can reproduce that level of theconsumption model of cloud.
(35:40):
But in an on-prem data center,man, they're just going to win
the game.
Like it's going to win the gamebecause it's impossible, it's
like so impossible but everybodywould benefit from it.
Like everybody would love that,because nobody's going to be
able to fully evacuate.
Unless you started in the cloud, you're probably never going to
truly leave all of your datacenters behind, right?
(36:01):
You're going to need somethingon prem.
So, yeah, man, I yeah.
Chris Miles (36:04):
Yeah, I think, I think it was very easy not easy,
but it was easier, at least inum the pre-cloud days to kind of
classify what type ofnetworking, like products and
things were out in the market,because there was always this
kind of delineation of likephysical location like like
branch routing, data centerrouting or data center switching
(36:25):
was.
It was a very different thing,right, that's like, even if I
just say that that impliescertain things about the
hardware campus and about it.
Nicolas Vibert (36:33):
What if its capabilities are?
Chris Miles (36:35):
know, right off the bat, right, whereas cloud has
just become this extension of IT, right, it's not?
It's like everything runs there, right.
So it's like it's incorporatingthe branch the way in, the like
.
Everything is there, right.
So it's just like so convoluted, yeah, yeah.
Nicolas Vibert (36:52):
So I think maybe we just need some new
vocabulary, more acronyms, yeah,yeah, but I think maybe we just
need some new vocabulary todescribe it More.
Acronyms yeah, probably.
But yeah, we need to separateit because I think it just
becomes very complicated andthen I end up having to answer a
(37:12):
questionnaire from vendors andthe questions that maybe apply
to me, to my products, don'tapply at all to you know.
I don't know like a restartWi-Fi or whoever.
Chris Miles (37:24):
So it's Gartner's problem, basically.
Tim McConnaughy (37:28):
I think that's true, though I mean speaking of
someone who's also had to fillout one of those 40-page reports
like there are.
Definitely sometimes they'llask a question or be like
explain your capability, andyou're just sitting there like
scratching your head, liketapping the pen on the desk,
like huh, I got to put somethinghere, but like our product
really doesn't play in thisparticular space.
Nicolas Vibert (37:48):
Yeah, well, one we.
We had a question recentlywhich was all your your product
doesn't support like drag anddrop and your product?
Tim McConnaughy (37:55):
doesn't support drag and drop.
You're like no, not exactly.
Nicolas Vibert (37:59):
What?
Because maybe there will besome products where you can do
some kind of Right like a policycreation in a UI or something.
Chris Miles (38:09):
Yeah, a map type thing, yeah.
Nicolas Vibert (38:13):
I'm not sure that, as a decision maker, we
think, oh, I won't buy thisproduct.
Chris Miles (38:20):
Gotta have that drag and drop man.
Tim McConnaughy (38:23):
Yeah, I can't build rules if I can't drag and
drop things in my UI.
That doesn't exist.
Nicolas Vibert (38:29):
Yeah.
Tim McConnaughy (38:29):
Now that's really interesting, I don't know
.
So I know we're running out oftime, but I don't know if we got
to an answer and I don't thinkthe three of us are
authoritative enough to make ananswer for everyone, anyway, but
even for ourselves.
It's very convoluted, it's veryephemeral, it's changing all the
(38:50):
fucking time.
Yeah, it's almost like and thisis the problem with a lot of
things, right, not justnetworking, but also security,
as we pointed out, it's veryreductive, right?
So, in order to get to the bareminimum the MVP definition, it
becomes so reductive that youcan almost sweep everything
underneath it, which is whatyou're trying not to do, right,
(39:11):
by defining it.
Nicolas Vibert (39:13):
Yeah, I won't even even try.
I enjoy the, the, actually thediversity, uh of of cloud
networking and the whole.
Um, even a few years ago, theway I thought of the cloud
networking is different fromwhat I think of it now, and
maybe you know, in the same fiveyears time, like, I guess what,
what do you think like cloudnetworking will look like?
(39:34):
You know, and can you answerthis without mentioning ai?
Chris Miles (39:38):
no, no, that's as far.
We got this far without sayingai, that's great yeah well.
Tim McConnaughy (39:45):
So I mean, but yeah, we're not gonna.
I don't think, I I don't evenknow right, like it's almost
like a uh, what's the word?
What I'm looking for like theuh singularity, like the
technical singularity, like liesfive years in the future, now
that AI is everybody's fricking,finding new and new, newer and
newer things to do with AI andwe're getting closer and closer
to to Skynet or whatever.
Nicolas Vibert (40:06):
Like who can even say and that's something we
have now seen in the pastcouple of years is in in vendor
analysis asking does yourproduct support some form of AI?
And I always like to kind ofdistinguish between is it about
AI for networking, do weleverage some form of AI to
(40:26):
optimize network traffic,predict patterns, or is it about
making the network more robust,quicker for AI workloads?
And I think both of them apply.
I think both are requirementswe're going to see more and more
from not just kind of cloudnetworking but broader
(40:49):
networking products is can yournetwork cater for AI workloads
and does your network leverageAI to provide better performance
?
Tim McConnaughy (40:58):
Right, yeah, so this is interesting because the
CSPs are like whole hog onbasically developing their own
silicon, which they've beendoing for a while, but
specifically for AI.
They don't want to be beholdento NVIDIA or anybody for GPUs a
while, but specifically for AI,like they don't want to be
beholden to NVIDIA or any of theyou know any anybody for GPUs
(41:22):
and GPU.
So doesn't that also come witha requirement then to and maybe
this is why all these wereported on the news a few weeks
ago that you know Microsoft isplanning to spend what?
A hundred billion dollars ontheir, on their upgrading their
infrastructure, and like AWSsaid 75 and Google said like 80
or something like that, do youthink a lot of that is going to
build the networking for AI,meaning networking to support AI
(41:43):
workloads?
Do you, you know cause peoplemay not be building their own,
or do you think we'll see areduction, like with deep?
You know this, the deep seekthing was like 50, 50, which is
what I kind of expected, whichlike, yeah, there are things
that are good, but also a lot ofit was kind of snowed over.
It was a snow job of buryingthe lead on a lot of the costs
and stuff, but it stillrepresents a movement forward
(42:06):
towards making it more.
What's the word I'm looking for?
Affordable, essentially to doAI, so do you think they'll just
kind of meet in the middlesomewhere?
What do you think we're goingto see first?
Nicolas Vibert (42:18):
I am equally as excited and dubious about AI for
networking.
I guess for a couple of reasons.
One is I wrote a blog postabout machine learning for
networks 15 years ago.
I wrote a blog post aboutmachine learning for networks 15
years ago.
I remember being I got thechance to sit in a room with
like a Cisco fellow which islike the highest engineering
(42:42):
level at Cisco and he waspresenting some machine learning
capabilities that he wasworking on at the time.
That was 2000.
It was more like maybe 2015,but it was like it was maybe 10
years ago.
But it was, like you know, way,way, way before we you know, we
yeah.
(43:04):
And I got very excited about thepossibilities, but nothing from
what I know kind of came out ofit.
Right, it was, maybe it wasmaybe it was too early, but
clearly, you know, we've seenlike some amazing advancements
and I'm a more than daily userof ChatGPT, but I still have
some I really want to see likereally applicable use cases.
(43:30):
Like show me okay, networkinghas been able to help us make a
better decision for this packet,right?
Yeah, it was supposed to go andfollow the you know the rising
table and I could not hang on.
we'll just go and send it toelsewhere because of you know
conditions, whatever that is,yeah, yeah, um, yeah yeah, just
(43:51):
just I want to see some moreevidence, but I'm, I'm, I'm
excited and I think just it'sgoing to really help us make,
create better products, help usdevelop our software faster, for
sure.
But we still want to see a bitmore before.
Let's say, I trust AI toenforce some security policies
(44:15):
for me, for example.
Tim McConnaughy (44:20):
That's the one.
Who was it?
It was John.
We had John Capobianco on there.
We were asking him the samething about.
You know, when is it going tobe the point where somebody is
going to trust this agent likean AI agent enough to to
actually go do network-levelchanges that could just blow up
the whole network?
Like, where's that?
Where do we get to that partwhere somebody's okay with
(44:40):
pressing that button and walkingaway?
You know, essentially so Ithink I'm with you.
I think it's going to still bea long time before we get to
that, no matter what, right.
Chris Miles (44:51):
We're too big of control freaks for that at this
point.
Tim McConnaughy (44:56):
Well, network engineers have to be, because
those five nines of uptime areon us.
Alright.
Well, we'll go ahead and wrapit there.
Nico, as always, it's awesomehaving you on and I'm really
glad we could do it.
We'll have to do it again inthe future, any time any final
thoughts there, chris, before weroll up.
Chris Miles (45:15):
No, I was just going to say Nico.
Where can people find youonline?
Nicolas Vibert (45:19):
Yeah, so you can find me on LinkedIn Again.
That's probably where I'm mostactive.
So LinkedIncom slash NicolasViver.
Tim McConnaughy (45:34):
We'll get it in the show notes too.
Yeah, yeah, yeah, all right.
Nicolas Vibert (45:37):
Good night people.
Tim McConnaughy (45:39):
All right, everybody.
Well, I don't know if we got tothe exact answer, but hopefully
somebody found some usefulinformation in all of this.
As always, I'm Tim, this isChris, our guest, nico, and
thanks for listening or watching, and we'll see you next time.
Hello and welcome to another episode of
the Cables to Clouds podcast.
I'm your host this week, tim,and with me, as always, is my
co-host, chris.
I'm at Carpe DMVPN on Blue Skyand Chris is at BGP Main on Blue
Sky, if I could speak correctly.
And with us is a returningguest, a good friend of the
(00:36):
podcast.
We're lucky to have him.
Nico Weber, would you like to?
Actually, I don't remember,nico, are you on?
I don't remember.
Are you on Blue Sky yet, nico?
Are you?
You to actually?
Nicolas Vibert (00:44):
I don't remember .
Nico, are you on Blue Sky yet?
Nico are you?
You don't do the social media.
No Well, I used to be onTwitter and pretty active, but I
left that, yeah.
Tim McConnaughy (00:52):
That was a good move.
Nicolas Vibert (00:55):
Pretty active on LinkedIn with videos and silly
blog posts, but that's whereyou'll find me.
Tim McConnaughy (01:01):
Yep, yep, yep, excellent.
And just in case, so we've hadyou on the podcast before.
It was an excellent episodetalking about Kubernetes and
Cilium, but for the people whomay not have heard or seen that
episode, go ahead and quicklyreintroduce yourself.
Nicolas Vibert (01:17):
Sure.
So I'm a senior staff technicalmarketing engineer at
iSurveillance slash Cisco and mybackground is network
engineering.
I've been in network for about20 years now.
I did my CCNA, ccnp, ccie backin the days.
I was CCIE for about 10 yearsuntil I let it go and worked for
(01:39):
Cisco for a while and then Iwas a boomerang back in through
the acquisition of iSurveillanceby Cisco last year.
So yeah, just happy to be backon this podcast.
Tim McConnaughy (01:52):
Yeah, we're definitely happy to have you
back and actually you came upwith a really cool topic and
that's why we brought you back.
And let me see if I can readthis correctly, just make sure
I'm getting it right.
So we were talking on LinkedIna little bit and you were like
you know, what I really love totackle was this idea of the
identity crisis of cloudnetworking, specifically, what's
cloud networking and what it is, what it isn't.
(02:14):
And I just realized that, likeyou know, there's so many
opinions on this that, like it'sa great topic and I'm sure if
you talk to 10, five engineers,you'll get 10 different opinions
, right, so I love it.
So let's just kick it off there, like what made you think of
this topic?
And like because I mean, sothis comes from somewhere.
You've got opinions on this,yeah.
Nicolas Vibert (02:41):
Well, I guess, as a marketing, technical,
marketing engineer, one of myresponsibilities is to work with
analysts, you know, like thelikes of Gardner and Forrester,
that kind of people, and answerquestionnaires about our product
and where we fit, and obviouslylet them, you know, rank
stackers against other companies, you know, in the quadrant
right.
And I was recently answeringone such query about cloud
(03:04):
networking.
I looked at all the differentvendors and I was like, hang on,
are these guys cloud networking?
And you just look at the listof people we were being compared
with, it was like apples andoranges, lemons, I don't know.
It was just a bunch ofdifferent companies and I just
kept thinking what is even cloudnetworking?
And you guys, you've had likecloud networking and you guys,
(03:24):
like you know, you've had acloud networking podcast for a
couple of years now a coupleyears, yep yeah, um, you
probably, even between yourself,have a different definition of
what cloud networking might be,so no, it's.
Tim McConnaughy (03:38):
It's funny to hear you say this because since
I became a TME, one of thethings I probably hate most are
those 40-page reportquestionnaires about hey, here's
, fill out 40 pages of why youfit into this quadrant and how
you're better, and you know,usually they're like, oh, don't
talk about your competitors inhere.
But then you get to see, likeyou said, like you go look at
(03:59):
the magic, the quote, unquotemagic quadrant or whatever.
They've all got their ownversion of stack ranking vendors
and you're looking at them likehuh, Like why am I?
These guys are like SSE orsomething.
They got nothing to do withwhat I do.
Right, so absolutely All right.
Nicolas Vibert (04:16):
So I'll ask a question then who do you think
was on this cloud networking,you know?
Who else we have ISOVANON,which, as I said, is now part of
Cisco.
There might be some that youthink, oh yeah, of course that's
a cloud networking company, andthere might be some that was
like, hang on, so give me your,give me your, give me your
(04:36):
guesses.
Chris Miles (04:37):
Like versus like versus, who is versus who isn't.
Nicolas Vibert (04:39):
Well, who do you think is on it?
Tim McConnaughy (04:41):
There's a list.
Nicolas Vibert (04:45):
Who would be on the list?
Yeah, I've got the list.
I'll tell you if you're.
Chris Miles (04:47):
If you're correct, or not nice uh yeah, well, I
mean, I would hope.
I would hope our currentemployer is on there as well.
Tim McConnaughy (04:53):
Uh, I'm sure a matrix has got to be on there
somewhere.
Yes, uh, I mean so obviously,alkira would I.
I mean I say obviously, but I Ibelieve that alkira would would
qualify and I feel like we'relike passing judgment on that.
I feel like an asshole, we'rejust passing judgment on these
companies.
But no, I mean I think from afocused cloud networking
perspective.
Yeah, I mean I think Alkirawould fit the bill for that.
(05:14):
Prosimo probably did before itwent under right.
I mean obviously Isovalent.
Did Calico's own?
Is it Tigera's the company thatactually does Calico?
Is that right?
Nicolas Vibert (05:30):
Yeah, they're actually not on it, but I
suppose they could be, becausethey offer a similar product to
what we do.
Tim McConnaughy (05:35):
Yeah, well, okay.
Now, this is, of course,because I only just started
getting into the Kubernetesstuff within the last few weeks
and I went straight to Ciliumbecause I already knew you and I
was very familiar with the labsand everything that you pointed
me towards.
But then I was readingsomething else and I was like,
oh, I should take a look atCalico and it's not the same,
and I don't want to compare thetwo products.
(05:55):
But I was like, just from a,you know, it's a CNI plugin,
right, like it fits at least theKubernetes CN, kubernetes, cni,
I that's.
And again, this gets into thewhole cloud networking thing.
Well, which you know, it'skubernetes, kubernetes
networking, which could berunning in the cloud.
So, yeah, you can see how itautomatically.
Nicolas Vibert (06:10):
Already we're starting to draw lines, right,
right yeah, and you know, like Ithink if I, if I had to guess,
I would have said aviatrix andprosimo, now kira, because
they're kind of for me likeobvious networking companies, um
, but then and then I saw, okay,I saw cisco and arista, juniper
(06:31):
.
Fair enough makes sense.
Tim McConnaughy (06:33):
So f5 yeah, that's yeah, are we?
That's like a fair point, right?
They route packets but their,their actual value, prop and
application is in the loadbalancing, right.
So that's a fair point.
Uh, oh no, didn't f5 boughtsomebody though?
Uh, a few years ago, um, whatwas the name of that company?
They bought a they bought acloud.
Chris Miles (06:53):
Yeah, was it engine x.
Nicolas Vibert (06:54):
Sorry, it's a bit of gen x yeah yeah, I'll
have to look it up.
Tim McConnaughy (06:58):
They bought an actual company that they were
like, but then they kind ofburied it.
They never came out with theproduct, the the.
They actually had aquote-unquote multi-cloud
product and they never came outwith it.
Nicolas Vibert (07:07):
Now.
I have to go find it Well.
Tim McConnaughy (07:08):
Cloudflare actually just today had a press
release about a new MCNS product.
Did you see it?
Oh, did I?
Okay.
Chris Miles (07:16):
Yeah, they acquired a company called Nefeli I think
it was about a year ago, rightand now it's being called ago,
right, um, and now it's beingcalled like of course they've
dubbed their, their brandedmagic, uh, moniker onto the
front of it now, but it justlike that seems like an
orchestration thing.
So but like I think, as we openthis up, that kind of speaks to
your point, nico, like what iscloud networking?
(07:38):
Like?
We have like companies justdoing maybe like kubernetes
networking and then maybemulti-cloud networking, and then
you know something as simple asjust orchestrating the native
constructs, like is that a cloudnetworking company?
Like it's uh, yeah, the, themoney.
Nicolas Vibert (07:53):
the waters are very muddy at this point yeah,
you know, I see, like the likesof you know broadcom vmware is
that.
You know, do they includevmware and I make vmware and
yeah, you know, like an sx orsomething.
Yeah, I'm assuming that's an sxright I would think so yeah,
but like would you?
would you describe an sx ascloud networking?
(08:16):
And you know, and again, Iworked on an sx and which is I?
I mean an sx couldn't be moredifferent than you know, like
what, what, what I don't know,f5 or prosimo, or if hvc says
right.
So this is what I mean about,like this, cloud networking
products and companies thatactually do very different
things.
Chris Miles (08:35):
Yeah, yeah, I feel like.
I feel like this is almostsomewhat adjacent to like the
kind of the evolution of wheresd went.
Oh yeah, we had a few companiesstart out that were obviously
pure SD-WAN companies, like youknow, kind of in that kind of
founding sector, so to say.
Tim McConnaughy (08:52):
Like Viptala Of the product.
Chris Miles (08:54):
Versa, versa, yeah, and there's one other one
Avello Cloud.
Tim McConnaughy (09:00):
Yeah, Avello.
Chris Miles (09:01):
And then, but you know, then we started to see the
likes of the major networkingvendors.
Uh, just start like maybe ormaybe, like I saw, mainly
networking vendors, and like wayin optimization they were just
yeah the way in ops too, yeah,firewall vendors, firewall
vendors, they would add like onefeature and now it's like okay,
well, now this is sd-wan it'slike I don't know I don't know
(09:22):
if you can really you, you knowput your merit on that Like
that's just like.
Nicolas Vibert (09:26):
Yeah, and it was like it went from like being
SD1 to being SASE to being ZTNA.
Tim McConnaughy (09:33):
you know it's so quickly.
Nicolas Vibert (09:36):
But we were like yeah, like sorry, it was like
even SD1, like, is that cloudnetworking?
Because you know, you used tothink about SD-WAN and you had,
like, you are your branch deviceand you would say, oh, let's go
and take all that traffic toOffice 365.
And instead of backhauling backto my data center and then
through my internet pipe, I sendit straight to my, you know,
(09:56):
over my local internet link.
That's some form of kind ofcloud networking as well.
Tim McConnaughy (10:01):
Not, you know, not the yeah, well, this is
interesting too, because the uhthis is where we get into some
other things where there's kindof a long-standing curiosity
about like is saz cloud the samelike?
You know, like when, uh, wetalk about this all the time,
where, like customers are like,or microsoft is like, we have,
uh, you know, so many millioncloud customers and it's like,
(10:21):
well, I mean they're, they'reoffice 365 customers.
I don't know if I would callthem cloud customers.
Nicolas Vibert (10:24):
And it's like well, I mean they're Office 365
customers.
I don't know if I would callthem cloud customers.
Tim McConnaughy (10:27):
Sure, it runs on Azure, but I don't know if
that makes them Azure customers.
You know what I mean.
Nicolas Vibert (10:32):
Like it's not quite the same, Let me ask you
guys, like where do you thinkcloud networking was invented?
Tim McConnaughy (10:41):
I mean like the CSPs.
Chris Miles (10:43):
Yeah, I mean mean ultimately, I think the.
If we look at the evolution ofcloud, you know kind of start in
the early days with, you know,launch of S3, and then you know,
eventually more things startedor like using cloud and needed
the natural progression andeventually things needed to be,
you know, able to talk to eachother, right, I don't remember
what was it.
When did VPC launch?
Was like 2010.
Wow, literally.
So that's to be you know, ableto talk to each other, right and
(11:04):
um, I don't remember what wasit.
When did vpc launch?
Was like 2010, 2000.
Well, literally.
Nicolas Vibert (11:08):
So that's to 25th of august 2009.
I found jeff biles uh,announced some blog posts.
Oh, nice, nice, um, and it'sjust, you know, you.
You read it and it's like, wow,you know.
That's like close to 16 yearsago and all still relevant,
well-written, clear.
The blog post was about, youknow, being able to create a
(11:31):
virtual isolated networkconnecting back to on-prem with
a VPN, so that for me that feelslike the origins and I'm sure
some people will disagree, butit feels like the origins of,
like, cloud networking.
Tim McConnaughy (11:45):
Well, here's the question, though Hold on
real quick, because we got to.
Some people will disagree, butit feels like the origins of
like cloud networking.
Well, here's the question,though Hold on real quick, cause
we gotta.
Nicolas Vibert (11:53):
I'm waiting for the data center people to start
screaming that like privatecloud.
What about private cloud?
Yeah, and you think okay.
Tim McConnaughy (12:03):
I think the response to this wasn't, you
know, things like OpenStack.
Yeah, that's a very good point.
Actually, I like OpenStackright.
Yeah, that's a very good point.
Actually, I think OpenStackwould probably be, because, I
mean, we're really talking abouta model, we're not just talking
about private cloud as a datacenter.
That's not you know its ownkind of abstracted workflow.
Because to get to that onprivate, yeah, I think the first
thing you would have been ableto work with to do that would
have been, yeah, openstack, totry to abstract away the compute
(12:26):
and stuff.
Nicolas Vibert (12:27):
And they had the Neutron or Quantum plugin which
was their way of creatingvirtual network and again I
think that's my recollectionwhere it was in response to the
cloud providers and taking offand they wanted to cloudify the
way networking was beingconsumed, on-prem right To offer
a similar kind of experience.
(12:48):
Yeah, it's again.
We still haven't really got tothe definition of cloud
networking, right?
Tim McConnaughy (12:54):
No, no, no, no, we'll get there.
We'll get there around 55minutes or 60.
Chris Miles (13:01):
Well, it's funny because I think cloud has always
been used pretty much as anabstraction.
Right At the end of the day,there's still kind of, you know,
hardware and components runningunder the hood that we all are
familiar with.
But I feel like, yeah, thecloud initially was supposed to
be this term.
That was like somewhat agnosticand you know, we had private
(13:21):
cloud, we have public cloud, but, like, private cloud is a very,
very different experience thanthan public cloud and I feel,
like the terminology has kind ofjust shifted cloud I mean.
In my mind cloud means publiccloud, um, and what that
ultimately translates to is, youknow, basically using, uh,
vendor or service providerspecific APIs to interact with
(13:44):
their environment to get it todo what you want to do.
So it's like I mean, that'swhat it sounds like to me.
I mean, ultimately, you'realways going to be able to do
what they give you access toright.
Yeah, just like MSP, right,there's no super level of
customization that you could dowith a private cloud Right.
So I mean, that's the way ittranslates to me.
Nicolas Vibert (14:05):
And I think that for me, cloud networking, it
does involve some form of publiccloud, because otherwise the
definition just becomes toobroad.
Yeah, it's too generic.
Chris Miles (14:18):
Yeah, Otherwise it's just networking.
Really, yeah, networking.
Nicolas Vibert (14:25):
But I think what you said around consuming the
MSPs APIs, I feel like the firstkind of real form of kind of
cloud networking companies, likethe likes of Aviatrix, was
really about like building a gapthat the cloud networking, the
MSPs where the networkingconstructs had some limitations.
(14:48):
So the startup was like, oh,let's go and create a product
that fits this gap.
That's fair.
Tim McConnaughy (14:54):
Yeah, no, I couldn't agree more, 100%, 100%
yeah, it's.
In fact, if you look at thehistory of our company, if you
look at the features we rolledout, first stopgap, fill in the
blank.
Like these don't exist in thehyperscaler and I think there
will always be a limitation towhat the hyperscalers can offer
to customers.
Just by the fact that they'rehyperscalers, right, they can't
(15:14):
give you the same granular levelof control that you can give to
if you're going to, if theygive it to you, they got to get
to everybody, right.
Chris Miles (15:20):
So that's yeah, because it's a service, right,
like, like hardware, likehardware vendor can give you
something off the shelf thatdoes everything under the hood,
because you control all of thatand you know it's probably in
your isolated environment, right.
But if you're providing aservice, there's only so much
you can do, um, for, uh, youknow, like with r&d development,
things like that, it just takesso much more work to get to
(15:40):
that, get to that level I thinkevery uh, even every msb, has a
different stance and culturetowards networking.
Nicolas Vibert (15:49):
I mean, I feel like AWS was, like you know, so
developer-centric and I don'tknow, google is, like, so proud
of its networking, so proud ofits backbone and, you know,
happy to share.
Oh look, we've got cables, youknow, under the Pacific Ocean,
the pacific oceans.
Look at that, it's amazing.
And like they're just very,very proud of someone's
networking and their pipes andtheir innovation around this.
(16:12):
Uh, they take now, that'sinteresting yeah, they have a
different perspective, I think.
I think that's a culturalaspect to uh, to to the
importance of networking yeahthe like.
Chris Miles (16:23):
it's always cool when they show those things, but
in my mind, whenever I see youknow, um, you know a service
provider, specifically a cloudservice provider, kind of toting
about like what kind ofconnectivity they've built with
you know undersea cables, allthis kind of stuff, all that
says to me at the end of the dayis like, there's capacity, like
the like.
Capacity will probably never bemy problem, which is good.
(16:45):
It doesn't change how I buildnetworks and I'll be honest with
you in the long run, the leastinteresting thing well, not the
least, but one of the most leastinteresting things to me in
networking has been QoS, and ifI don't need that anymore, then
I'm very happy with that.
Nicolas Vibert (17:01):
So I think this is progression for sure.
Chris Miles (17:05):
But I don't know.
Like yeah, networking to melike the undersea cables that's
all cool, the CDNs that's coolas well, but like I'll be honest
like that doesn't move theneedle for me, like in terms of
cool technology.
Tim McConnaughy (17:18):
So it's interesting, nico, you have the
observation that Google cares alot about its networking,
because I think that's true.
What you said about is buildingthe undersea cables.
I think all that is true, but,like, when I look at google's,
the way that you as a customerhave to consume google's
networking, to me it seems likeextremely developer centric.
Like like don't even theconcept of global vpc and
(17:41):
specifically the idea that likewe can just do it all for you.
It seems like they really justwant you to not care about
networking, or just for meworking with it.
It seems, you know, like theauto building uh vpcs where
they'll auto do all your subnetsfor you.
It seemed to me like they theywere focused on more.
So it's interesting that theopinion that you have that is
different yeah, it's.
Nicolas Vibert (18:02):
I mean to be fair, like I um, I was at cisco
last week and I was showing ademo with a couple of colleagues
that uh, which was like ummeshing a couple of capabilities
cluster one was on-prem, onewas um in, uh, the cloud and we
picked, we picked uh, gk, uh,because we thought actually from
(18:23):
a networking perspective, thatwas the easiest.
Tim McConnaughy (18:25):
Oh, for sure, I think so.
With the global VPC andeverything and the way they do
it.
I think GKE is the easiest.
Nicolas Vibert (18:32):
And you know, doing some cool connecting
machine clusters together, doingsome low balancing across sites
.
That's a nice form of cloudnetworking.
Tim McConnaughy (18:41):
Yeah.
Nicolas Vibert (18:42):
It's interesting .
Tim McConnaughy (18:42):
I'm actually just getting into this myself
because we're kind of doing somestuff with Kubernetes as well,
so I've been so late, Like Iknow we had you on what like
almost a year ago now orsomething like that.
It's been a while and I was oneof those things where I was
like I'll get around to it.
I didn't really get a chanceuntil a few weeks ago, but the
more I learned about it, themore interesting it is,
especially the networking aspectof it about it, the more
(19:05):
interesting it is, especiallythe networking aspect of it,
even though it's like 90% of itis abstracted away, at least
within the cluster, certainlywithin the cluster it is.
Nicolas Vibert (19:11):
Which is interesting from my perspective.
It was my first time presentingat Cisco Live.
I've been to Cisco Live manytimes but I've been presenting
at KubeCon, for example yeah,and presenting our networking
platform for Kubernetes.
It's very different when you goto somewhere like Cisco Live.
(19:32):
Compared to KubeCon, audienceis very different.
Yeah, exactly, very, exactly,very different audience and,
like some of the KubeCon,they're more interested in like,
what can you do with all thisnetworking?
Like you know more about theapplication stuff, can you do
things like canary testing andload balancing and introducing
new version of my app, do someredirection?
(19:54):
Lots of different, more, moreagain application-centric,
whereas I go to Cisco Live andwhat the first question I get is
like, show me your packet work,show me the headers, and you
know, show me BGP, you know, andit's just like oh boy, but so,
yeah, it's kind of again adifferent culture.
But what was?
(20:14):
I guess?
What I found fascinating is theplatform and Kubernetes.
Networking is aboutcloud-native applications, which
is mainly containers and mainlymanaged and orchestrated by
Kubernetes, and I think that'sone of the reasons we were
invited in this vendor's analystreport is because we do
(20:35):
networking for cloud-n nativeapplications.
So does it qualify?
As you know, is that arequirement to be cloud
networking?
Do you need to be able tosupport modern containerized
applications, or is that just akind of a side aspect of it?
That's a great question.
Chris Miles (20:53):
Yeah, I mean, I think like we're at this point
where we probably like I feellike we don't want to fall like
in the same direction, likemaybe security has where there's
, you know, there's so manydifferent categories for things,
like there's, you know like somany acronyms, cnap, all this
kind of stuff.
There's probably been twoannounced today as yeah, there's
(21:15):
a lot um but like I don't knowif we wanted like venture into
that territory where we havelike ultimate levels of
classifications to every one ofthese things, like um, I mean,
we're already there, we cangenerally talk about it, but
there's just no, you know, likeofficial, you know, once gartner
does it, then then then it'sreal right um, but yeah, I don't
know if I wanted to go that,that route, but I mean,
ultimately there there isconfusion.
(21:36):
So you know, the consumer at theend of the day, is going to be
the one that's confused whenthey hear this term.
Right, so, um, it might benecessary, I don't know.
Tim McConnaughy (21:44):
So, so this, uh , you know you get me thinking,
though, because, like Kubernetes, the whole thing with
Kubernetes is that we, you knowyou can have this cloud native,
Cloud native and Kubernetes likego together.
Right, the CNCF owns the wholething.
Like there's so much it'sjoined at the hip.
Right, we say cloud native,cloud native, cloud native.
You kind of wondered, you know.
Now we have to like, okay, whatthe hell does cloud native mean
(22:06):
?
In order to get to cloud native, what the hell is cloud native?
We had a discussion on this aswell we did, we did a while back
.
Chris Miles (22:13):
Yeah, it's like I think it's strayed from what it
what it used to mean or what itwas intended to mean as you can
force.
Tim McConnaughy (22:19):
You can run a kubernetes cluster on prem right
or, you know, in a middle mileprovider or something.
I don't think I'm not, I don'tknow how many people do that.
To be honest with you, I thinka lot of people consume the, the
managed kubernetes.
Kubernetes is prettycomplicated, but well, actually
I mean the.
Nicolas Vibert (22:33):
I mean just again talking to customers at
Cisco Live and KubeCon.
So many are doing self-manageand you know, very often on the
like, you know like OpenShift orRancher, and that's why I had
so many conversations aroundthis last week at Cisco Live.
Again, it's about like okay,I'm building my own Kubernetes
(22:54):
cluster, managing my own on-prem, how do I connect to my Cisco
ACI, for example?
Yeah, of course, any otherfabric.
Tim McConnaughy (23:06):
Just outside the cluster.
How are we making connectivityright?
Yeah, how do we expose our?
Nicolas Vibert (23:10):
applications applications.
What's the best practices?
Again around BGP and Ingressand make our cloud native
applications accessible from theoutside.
Tim McConnaughy (23:25):
Yeah, it's funny because with Kubernetes,
specifically for Ingressspecifically, you have to build,
you need a load balancerservice, something for the
ingress to the service you'vedone.
So now a load balancer nowbecomes part of cloud networking
, Like all these services.
These extra services end upbeing part of cloud networking.
(23:46):
Right, it's not just networkinganymore, there's all these
services included.
Nicolas Vibert (23:51):
Yeah, so is service mesh a service mesh or
an ingress controller?
Is that cloud networking?
Tim McConnaughy (23:59):
I think it has to be with Kubernetes
specifically?
I think it has to be right, theservice mesh especially.
You've got your sidecar.
You're building tunnels.
You're just making a tunnelmesh between clusters with your
sidecars.
Chris Miles (24:12):
That one almost ventures just into application
networking.
Nicolas Vibert (24:15):
to me, yeah, maybe At the end of the day, I
guess like what is that that's?
Chris Miles (24:20):
fair.
I mean it's turtles all the waydown right.
Eventually it's the same thing,but like I don't know, we're
probably splitting hairs.
Tim McConnaughy (24:28):
No, I mean, that's a fair point though
Application, because it's theapplication.
So yeah, I mean now that you'reasking yourself, god, this is
getting so crazy.
Nicolas Vibert (24:36):
So, philosophical.
Tim McConnaughy (24:38):
Yeah it's almost Socratic in the method,
right?
Because then you have to saywell, is the Envoy like for
Istio, is the Envoy anapplication or is it a
networking?
Nicolas Vibert (24:51):
device or is it both?
Yeah, yeah, yeah, I mean it'sproxy-wise.
So yeah, it's kind of.
You know, defining a categoryis just really hard when it just
keeps things keep evolving.
But I think for me, I guess,the characteristics of cloud
networking.
There are things that I thinkwe can say, okay, the one thing
that all have in common for meis that they're API-centric.
(25:13):
That would be, I guess, my ownpersonal take, and maybe you
know I'm sure there are alsocloud networking products that
are maybe more consumed througha UI, but I think the aspiration
should be that you consume yourcloud networking.
Your networks are API, builtthrough an API.
Tim McConnaughy (25:34):
Okay, okay, yeah, I think that's Well.
What about Terraform?
Did you consider that the samething?
It's IOC, right?
Nicolas Vibert (25:41):
Yeah, with an API, some form of automation, a
playbook, a Terraform module,okay, yeah, yeah, that should be
part of the goal when you startleveraging a cloud networking.
Part of the goal when you startleveraging a cloud networking
is that you template, create atemplate or automate it.
If you claim you have a cloudnetworking product and you don't
(26:07):
have an API, then you're lying.
Tim McConnaughy (26:12):
Or, at the very least, if your product is not
interacting with the cloudprovider APIs to do the building
, like that kind of stuff.
Yeah, I think that's definitelytrue.
Yeah, I think after I askedabout the Terraform question, I
realized that what youoriginally had meant when you
said that so yeah, it is APIsall the way down, right, that's
how you generally interact with,whether you're doing it on a
(26:34):
CLI, whether you're doing it ona CLI, whether you're doing it
with their SDK, whatever.
That is right.
Terraform, it's an API call atthe end of the day.
So you asked a question on thenotes here that I really wanted
to get to and I don't want topick out a specific company, but
, like you had said, essentiallyis Middle Mile a cloud
networking company, somebodythat provides connectivity to
(26:56):
the cloud?
Would we consider them a cloudnetworking company.
Nicolas Vibert (27:03):
You have an opinion, yeah, you could argue,
and again you could say thingslike Equinix, megapod.
They provide some direct linksto the cloud from on-prem to the
cloud.
I think they also enable you toconnect your clouds together,
so some form of inter-cloudconnectivity.
(27:27):
Now, I think because it's moreof a physical layer to and from
the cloud.
I don't think it necessarilyqualifies as such.
I guess in my head it's more.
I guess cloud network, in myhead, is more about creating
networks within a cloudenvironment that isn't consumed
(27:52):
by applications that live in thecloud and that can be virtual
machines, that can be bandwidthor that can be containers.
Chris Miles (27:59):
Okay, so real quick then I was going to say an
important thing in thatclassification is they also
provide the connectivity likedirect connectivity options for
like SaaS providers as well.
Right, so that's like servicesrunning on cloud, but maybe
you're not getting into aprivate environment that you
manage on top of cloud.
So it's like running on cloud,but maybe you're not, you know,
getting into a privateenvironment that you manage on
top of cloud.
So it's like, yeah, like whatdoes that translate to?
Tim McConnaughy (28:23):
Well, I also have to challenge a little bit
the idea that if Megaportconnects two clouds together but
it's physical, that makes themnot cloud.
Because that then question isso if I connect AWS to Azure and
I do it over the internetinstead of using, because I mean
, you know, there's no directconnect between AWS and Azure.
That's obviously right.
(28:43):
So is that cloud networking orit's going over the internet?
Are we counting the internet asa medium?
You know what I mean.
Chris Miles (28:51):
But if we're thinking about like this concept
of at least the CSPs providednetworking options, you're
basically building networkingacross their backbones.
Whatever services they'veexposed, they allow you to build
networking on their platform.
What is different about that toMegaport or to Equinix?
That's the exact same thing.
(29:12):
They have an underlying set ofhardware services, a backbone of
something.
But the way you interact withit is essentially the same.
You make an API call and yousay build this connectivity.
So how is it not cloudnetworking at that point?
Tim McConnaughy (29:26):
We're not saying you're wrong.
I'd like to justify a littlebit.
Nicolas Vibert (29:31):
I suppose it's.
You don't necessarily use theirnetworks to run applications.
Tim McConnaughy (29:39):
Okay, so it's that you're not building apps on
their network?
Nicolas Vibert (29:42):
Yeah, and maybe I don't know enough about what
they do.
I guess it almost feels like asubcategory of cloud networking.
Maybe this is how we kind ofbreak it down right yeah clam
cloud networking as a kind ofoverall category, and then
you've got the intercloud, andthen you have the I mean it's
(30:05):
just not easy, right likethere's a reason why everybody
struggles with it.
Tim McConnaughy (30:08):
This definition , these definitions, is not easy
.
I think it's very hard to todraw these lines and to figure
out what lines to draw.
Nicolas Vibert (30:15):
And you know we talked.
You were saying like Cloudflareor maybe like CDN I don't
necessarily Akamai.
Tim McConnaughy (30:22):
Yeah, that's a really good question, man.
That's a super good.
That's even harder than theMegaport one, by far, I think,
or Megaport Equinix Middle Milegenerally by far, because, you
know, the CDNs are technicallyusually attached to the cloud
backbone, but like they're not,so they're, they're delivering
content right.
That is, over a network that isusually cloud based in some way
(30:46):
, whether it be Cloudflare ordirectly connected like with the
AWS offers.
Man, I don't know what do youthink?
Chris?
That's a tough one.
Chris Miles (30:54):
Yeah, it is tough, but I mean that one one, that's
one that I feel like has beenaround in some shape or form for
a long time as well, like evenbefore cloud oh yeah, cdns, yeah
yeah well, like I mean, yeah,like I remember, like back in
the day we used to, I used towork at a big mpls provider and
like we got a lot of circuitsfrom level three and at level
three, had a lot of these kindof front front door scrubbing
(31:17):
services and things that arejust very adjacent to what those
providers do today andobviously they have a much
bigger footprint and things likethat.
So it's like I don't know Thenso does it?
If that's the case, then thecloud networking start back then
, like yeah yeah it is, it is,it is, it is, it is, it is.
Nicolas Vibert (31:35):
Yeah, yeah, it is, it is.
It is very hard to say like,yeah, what, what was?
What started it all?
Um, yeah, okay, so what aboutsecurity?
Then?
Like, do we is, let's say,consistent security across cloud
?
Is that a must?
Is that like um, because wecan't do talk, we can't talk
about cloud networking withouttalking about cloud networking.
(31:57):
Security Is encryption, isconsistent networking across the
multiple clouds?
Yeah, some form of requirementto say, okay, we do cloud
networking.
Tim McConnaughy (32:12):
So, yeah, obviously, we work for Aviatrix
and we literally this is kind ofthe mission statement of the
company.
So, to put the company aside, Ithink, even putting the company
aside, I think you know ifyou're gonna this whole shared
responsibility model that allproviders have, you know,
customers are still in charge oftheir own data, right, the data
(32:35):
that's runs across their cloudnetworks.
Now it's encrypted in the rack,essentially, or across the
backbone, or however you want tosay it on the cloud provider
side, but ultimately a customer,but that you know.
Then you look at things likesalt typhoon uh, the sulfide
typhoon attacks, uh, you knowowning telecom networks and
being able to just do man in themiddle, you know captures of
(32:56):
open, of unencrypted traffic.
So, yeah, I mean, I do think Ithink it's a requirement, in
that a customer would be veryfoolish not to run some of their
, you know, do it themselves.
I wouldn't trust the providerto secure all of my traffic, but
does it?
Is it quote unquote required?
Like from a technical RFCperspective, if you will?
(33:18):
I mean, obviously not right.
Like AWS supports GRE acrossyou build a GRE tunnel from a
TGW Connect, for example, and ofcourse you can't do GRE in
Azure, but that's because that'swhat they're using.
On the underlay right.
Chris Miles (33:32):
It's funny how many directions this can go in,
because that's the first thoughtyou had, tim, about this.
I call back to just when we'retalking about the different
architectures that we see withincloud and how it's a very
strong fundamental shift withwhat we've been doing for 20, 30
years at this point.
Tim McConnaughy (33:52):
Oh yeah, the Volus coupling model.
Chris Miles (33:53):
Yeah, well, not even that.
I'm just talking.
If we think of the concept of aglobal VPC, like that's one big
network that exists everywhere.
Like the difficulty toimplement something in there,
like zero trust, which is aprominent thing you know from
from a networking perspective.
Like doing that in a globalconstruct that is just all
meshed together at one point,like that becomes a conversation
(34:15):
.
That's very difficult to getthat in there, right?
Tim McConnaughy (34:19):
What's your opinion, Nico?
Nicolas Vibert (34:22):
I can't imagine doing like a global VPC back
when I used to be like a networkoperator.
I just imagine there's adifficulty of doing something
that is so well crafted that youknow like you you get, you know
you deploy VPC in GK and youdon't really care Like you, just
(34:46):
you know your applicationreceiver, it just works and you
know.
And that's just thinking again,thinking back to the days where
I had to connect multiple,multiple sites together and
having to worry about addressing.
And I think that's the beauty oflike a really well designed
cloud networking like GKE, butagain also, providers also do it
(35:07):
very well, but it's all theintelligence and the automation
under the hood to, you know,make it happen and being able to
replicate it on-prem is justabout impossible.
Tim McConnaughy (35:20):
We've said that for even when Alex was here,
that we used to talk about allthe time that the people that
are going to win the game aregoing to be yeah, I know who is
that?
I don't know who that is eitherthe people who can, whatever
company or companies finally areable to build the product that
can reproduce that level of theconsumption model of cloud.
(35:40):
But in an on-prem data center,man, they're just going to win
the game.
Like it's going to win the gamebecause it's impossible, it's
like so impossible but everybodywould benefit from it.
Like everybody would love that,because nobody's going to be
able to fully evacuate.
Unless you started in the cloud, you're probably never going to
truly leave all of your datacenters behind, right?
(36:01):
You're going to need somethingon prem.
So, yeah, man, I yeah.
Chris Miles (36:04):
Yeah, I think, I think it was very easy not easy,
but it was easier, at least inum the pre-cloud days to kind of
classify what type ofnetworking, like products and
things were out in the market,because there was always this
kind of delineation of likephysical location like like
branch routing, data centerrouting or data center switching
(36:25):
was.
It was a very different thing,right, that's like, even if I
just say that that impliescertain things about the
hardware campus and about it.
Nicolas Vibert (36:33):
What if its capabilities are?
Chris Miles (36:35):
know, right off the bat, right, whereas cloud has
just become this extension of IT, right, it's not?
It's like everything runs there, right.
So it's like it's incorporatingthe branch the way in, the like
.
Everything is there, right.
So it's just like so convoluted, yeah, yeah.
Nicolas Vibert (36:52):
So I think maybe we just need some new
vocabulary, more acronyms, yeah,yeah, but I think maybe we just
need some new vocabulary todescribe it More.
Acronyms yeah, probably.
But yeah, we need to separateit because I think it just
becomes very complicated andthen I end up having to answer a
(37:12):
questionnaire from vendors andthe questions that maybe apply
to me, to my products, don'tapply at all to you know.
I don't know like a restartWi-Fi or whoever.
Chris Miles (37:24):
So it's Gartner's problem, basically.
Tim McConnaughy (37:28):
I think that's true, though I mean speaking of
someone who's also had to fillout one of those 40-page reports
like there are.
Definitely sometimes they'llask a question or be like
explain your capability, andyou're just sitting there like
scratching your head, liketapping the pen on the desk,
like huh, I got to put somethinghere, but like our product
really doesn't play in thisparticular space.
Nicolas Vibert (37:48):
Yeah, well, one we.
We had a question recentlywhich was all your your product
doesn't support like drag anddrop and your product?
Tim McConnaughy (37:55):
doesn't support drag and drop.
You're like no, not exactly.
Nicolas Vibert (37:59):
What?
Because maybe there will besome products where you can do
some kind of Right like a policycreation in a UI or something.
Chris Miles (38:09):
Yeah, a map type thing, yeah.
Nicolas Vibert (38:13):
I'm not sure that, as a decision maker, we
think, oh, I won't buy thisproduct.
Chris Miles (38:20):
Gotta have that drag and drop man.
Tim McConnaughy (38:23):
Yeah, I can't build rules if I can't drag and
drop things in my UI.
That doesn't exist.
Nicolas Vibert (38:29):
Yeah.
Tim McConnaughy (38:29):
Now that's really interesting, I don't know
.
So I know we're running out oftime, but I don't know if we got
to an answer and I don't thinkthe three of us are
authoritative enough to make ananswer for everyone, anyway, but
even for ourselves.
It's very convoluted, it's veryephemeral, it's changing all the
(38:50):
fucking time.
Yeah, it's almost like and thisis the problem with a lot of
things, right, not justnetworking, but also security,
as we pointed out, it's veryreductive, right?
So, in order to get to the bareminimum the MVP definition, it
becomes so reductive that youcan almost sweep everything
underneath it, which is whatyou're trying not to do, right,
(39:11):
by defining it.
Nicolas Vibert (39:13):
Yeah, I won't even even try.
I enjoy the, the, actually thediversity, uh of of cloud
networking and the whole.
Um, even a few years ago, theway I thought of the cloud
networking is different fromwhat I think of it now, and
maybe you know, in the same fiveyears time, like, I guess what,
what do you think like cloudnetworking will look like?
(39:34):
You know, and can you answerthis without mentioning ai?
Chris Miles (39:38):
no, no, that's as far.
We got this far without sayingai, that's great yeah well.
Tim McConnaughy (39:45):
So I mean, but yeah, we're not gonna.
I don't think, I I don't evenknow right, like it's almost
like a uh, what's the word?
What I'm looking for like theuh singularity, like the
technical singularity, like liesfive years in the future, now
that AI is everybody's fricking,finding new and new, newer and
newer things to do with AI andwe're getting closer and closer
to to Skynet or whatever.
Nicolas Vibert (40:06):
Like who can even say and that's something we
have now seen in the pastcouple of years is in in vendor
analysis asking does yourproduct support some form of AI?
And I always like to kind ofdistinguish between is it about
AI for networking, do weleverage some form of AI to
(40:26):
optimize network traffic,predict patterns, or is it about
making the network more robust,quicker for AI workloads?
And I think both of them apply.
I think both are requirementswe're going to see more and more
from not just kind of cloudnetworking but broader
(40:49):
networking products is can yournetwork cater for AI workloads
and does your network leverageAI to provide better performance
?
Tim McConnaughy (40:58):
Right, yeah, so this is interesting because the
CSPs are like whole hog onbasically developing their own
silicon, which they've beendoing for a while, but
specifically for AI.
They don't want to be beholdento NVIDIA or anybody for GPUs a
while, but specifically for AI,like they don't want to be
beholden to NVIDIA or any of theyou know any anybody for GPUs
(41:22):
and GPU.
So doesn't that also come witha requirement then to and maybe
this is why all these wereported on the news a few weeks
ago that you know Microsoft isplanning to spend what?
A hundred billion dollars ontheir, on their upgrading their
infrastructure, and like AWSsaid 75 and Google said like 80
or something like that, do youthink a lot of that is going to
build the networking for AI,meaning networking to support AI
(41:43):
workloads?
Do you, you know cause peoplemay not be building their own,
or do you think we'll see areduction, like with deep?
You know this, the deep seekthing was like 50, 50, which is
what I kind of expected, whichlike, yeah, there are things
that are good, but also a lot ofit was kind of snowed over.
It was a snow job of buryingthe lead on a lot of the costs
and stuff, but it stillrepresents a movement forward
(42:06):
towards making it more.
What's the word I'm looking for?
Affordable, essentially to doAI, so do you think they'll just
kind of meet in the middlesomewhere?
What do you think we're goingto see first?
Nicolas Vibert (42:18):
I am equally as excited and dubious about AI for
networking.
I guess for a couple of reasons.
One is I wrote a blog postabout machine learning for
networks 15 years ago.
I wrote a blog post aboutmachine learning for networks 15
years ago.
I remember being I got thechance to sit in a room with
like a Cisco fellow which islike the highest engineering
(42:42):
level at Cisco and he waspresenting some machine learning
capabilities that he wasworking on at the time.
That was 2000.
It was more like maybe 2015,but it was like it was maybe 10
years ago.
But it was, like you know, way,way, way before we you know, we
yeah.
(43:04):
And I got very excited about thepossibilities, but nothing from
what I know kind of came out ofit.
Right, it was, maybe it wasmaybe it was too early, but
clearly, you know, we've seenlike some amazing advancements
and I'm a more than daily userof ChatGPT, but I still have
some I really want to see likereally applicable use cases.
(43:30):
Like show me okay, networkinghas been able to help us make a
better decision for this packet,right?
Yeah, it was supposed to go andfollow the you know the rising
table and I could not hang on.
we'll just go and send it toelsewhere because of you know
conditions, whatever that is,yeah, yeah, um, yeah yeah, just
(43:51):
just I want to see some moreevidence, but I'm, I'm, I'm
excited and I think just it'sgoing to really help us make,
create better products, help usdevelop our software faster, for
sure.
But we still want to see a bitmore before.
Let's say, I trust AI toenforce some security policies
(44:15):
for me, for example.
Tim McConnaughy (44:20):
That's the one.
Who was it?
It was John.
We had John Capobianco on there.
We were asking him the samething about.
You know, when is it going tobe the point where somebody is
going to trust this agent likean AI agent enough to to
actually go do network-levelchanges that could just blow up
the whole network?
Like, where's that?
Where do we get to that partwhere somebody's okay with
(44:40):
pressing that button and walkingaway?
You know, essentially so Ithink I'm with you.
I think it's going to still bea long time before we get to
that, no matter what, right.
Chris Miles (44:51):
We're too big of control freaks for that at this
point.
Tim McConnaughy (44:56):
Well, network engineers have to be, because
those five nines of uptime areon us.
Alright.
Well, we'll go ahead and wrapit there.
Nico, as always, it's awesomehaving you on and I'm really
glad we could do it.
We'll have to do it again inthe future, any time any final
thoughts there, chris, before weroll up.
Chris Miles (45:15):
No, I was just going to say Nico.
Where can people find youonline?
Nicolas Vibert (45:19):
Yeah, so you can find me on LinkedIn Again.
That's probably where I'm mostactive.
So LinkedIncom slash NicolasViver.
Tim McConnaughy (45:34):
We'll get it in the show notes too.
Yeah, yeah, yeah, all right.
Nicolas Vibert (45:37):
Good night people.
Tim McConnaughy (45:39):
All right, everybody.
Well, I don't know if we got tothe exact answer, but hopefully
somebody found some usefulinformation in all of this.
As always, I'm Tim, this isChris, our guest, nico, and
thanks for listening or watching, and we'll see you next time.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!