Layer5 Service Meshes, 5 Dec 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Adopting Network Service Mesh with Meshery

Description

Trying to understand the service mesh strata? Brace yourself. It gets a bit meshy. Data plane, control plane, and a… management plane. What is a management plane and how is it different than a control plane?

Learn how Meshery, an open source, multi-service mesh management plane integrates with Network Service Mesh as we discuss and demonstrate their complimentary architecture. As a multi-mesh management plane, Meshery manages the lifecycle of NSM, providing insight into its CNFs while facilitating benchmarking of various configuration scenarios of Network Service Mesh.

See how these three service mesh planes combine to facilitate performance comparisons of services (applications) on and off the mesh and across different meshes. Understand the caveats of onboarding your applications onto a service mesh as Meshery validates best practices configuration of NSM and its CNFs.

https://networkservicemesh.io/events/nsmcon2019#adopting-network-service-mesh-with-meshery

A

For those that do come up after us on stage just know there aren't there aren't cold hands in the room, so yeah, so you didn't hear me squeal earlier so anyway be prepared for some dad jokes and some other other corny things to come. So my name is Leah. This is a grecian we're stoked to be here.

A

We've been working in the NSM community for a little while now we've been fortunate enough to have a couple of other folks who were deeper into the NSM community than we are from Lumina networks, and so you may, you may know these faces from other presentations, but Harsanyi and Prem fortunate enough to have them come over and work within an open source project that we're stewarding will tell you all about that one and in no apologetic way we will be.

A

What was the there was an apology at the end of the last talk for Hawking, something we're gonna, but there's no apologies for Hawking when we're gonna rock this work, it's gonna, which is hopefully a bunch of hopefully helpful, helpful things. Speaking of helpful things. There is a couple, a couple of books that I was able to get out here, and one of them is free for those who wear the noob, hello, hello. My name is noob like some of us.

A

Do, though, the one on the left hand side is a short report, but it's it's still, even though it's about a year old, there's still an accurate introduction to the service mesh land and, if you're like me, it's somewhat helpful because it calls out it presents service meshes in context of some things. You might already know things like a container Orchestrator things like a client-side library things you might have used in lieu of a service mesh until they became our becoming popular anyway free stuff.

A

Also, there's a book signing or a couple of them on the sto one. So.

A

Yeah boy hadn't really thought about how to present this line other than just sort of another corny joke about it being messy out there there's a lot of this is just an amazing number of ways you can use mesh and sort of make new words out of it. The thing is, this is true.

A

It is a bit messy out there and as a new genre of technology comes forward so to do a number of new projects, a number of new products, some of whom will take that buzzword and service mash and sort of just rub some of that buzz enos on their product yeah, maybe in advance of their product, really being sort of a sanctified, blessed service mesh. But but you know who can blame them? Who doesn't? Who doesn't want to rub? Some rub some mesh on.

A

Who can tell me who can take a guess how many service meshes are out there? Who's got a who's got a number. It doesn't very good guess, a very good number, dozens, a great guess and in some respects depends upon how you count. As a matter of fact, there's we've been counting, there's a service, meshed landscape, it's layer, 5, io, / landscape that begins to count them up and it doesn't have all of them.

A

But but it's kind of interesting cuz it, because it's more than a dozen I would think that any one of you, if you were to walk around the conference and say you know, ask someone else, hey how many, how many service you know you've heard about service specialist? How many are there? They probably you, know prattle off like to some of us three of the more popular ones, some that are going to be talked about even more today, so kind of interesting. So the the point here is that the world is a multi mesh world.

A

It's my belief that.

A

Dissimilar from the container orchestration wars where, where we still have a number of container orchestrators, they're still being maintained, people are still using them, but but here in the tech community we kind of look down our our big, long noses and a bigoted way and and sort of chuckle. When someone says dr. sore, okay can't believe you said, you know that you admitted to still using that when actually it still serves, you know many many use cases and maybe all of someone's use cases all of an organization's use cases, and so I think it's.

A

What I'm trying to say is: let's be inclusive, I think it's okay to to use the to ride the horse that didn't necessarily win. That particular race point is the analog here. Is that from my perspective, there won't.

A

While we will have some convergence, there won't be dozens or 20 or so surface meshes a year or more from now that will, you know, consolidate a little bit there's this natural flow of how those things work, but that will still have some of these will still be here though it was, it will be a multi mesh world. Come talk to me about some of the particulars about that that off record we can out I'll, give you some insights as to maybe why that would continue to be because of this.

A

There are two independent initiatives to define abstract specifications for interfacing with multiple meshes, who was at cube Connie you this last six months ago. What have you very good? We are I am in good company. This is great service. Smi surface mesh interface was announced on stage.

A

There was some nice pre-recorded demos pre-recorded, because because because there was a bit of magic behind the scenes, because it was a young specification of the specification has advanced since then, and this specification SMI is oriented toward providing a single API surface for interfacing to any number of different service meshes. It's a young spec, so it's only gotten so far in its use cases, and it's only gotten so far in its compatibility with the world of many meshes out there here recently. Anybody at VMworld I think was there too that just happened.

A

The bean world us I guess is the anybody all right very good, and there was some some of us wearing VM VMware badges will probably add that and so announced. There also was at the time it wasn't called Hamlet, but the internal project name for this. Second complementary vendor-neutral specification about how to interface agnostically with service meshes is called Hamlet. It's by-and-large stewarded by VMware, but there are others Google and others participating, and it is for purposes of helping you federated together. So maybe you've got hetero.

A

Homeo genius meshes it's the same type of mesh, but in different clusters, different places, some meshes have capabilities and facilitate for multi cluster mesh or like bringing together those those clusters and running a single, overarching mesh or running them in a highly resilient way.

A

Hamlet is a specification for for that and for bringing together heterogeneous message, meshes I'm going to say that word so many times, I mess myself up here. Fine anyway, is to federate them and have them exchange in and out services across one another. It's very interesting spec we're participating in that as well, and hence the project that we've been putting in a fair bit of time into. If I didn't say the word mesh enough. Here's another so turns out like a mesh. Urea is what we've called it: it's a Multi mesh manager.

A

It is what we consider a management plane.

A

We're proud of it thus far. There's a decent community. That's formed around it, a lot of a lot of students coming to cut their teeth on open source on this project. So it's it's pretty neat. The reason that this is important is that it is a multi mesh world. As a matter fact, it was a google Summer of Code project as well. This last summer, that was nice. It goes sets off two interface and advance both of these specifications.

A

Again going back to that belief that we will be in this multi mesh world for some time, it doesn't necessarily take a bunch of insight to sort of look at the vendors behind these various meshes and their competing interests, also to look at which meshes are governed by vendor neutral bodies like the those that the Linux Foundation has like the scenes yet and those that aren't and it's because of those because of those things that you know I think will continue to live in a multi mesh world, and so let me let's say: let's show you what mesh sure he looks like I'm the interface.

B

All right so I already have.

C

The page up so here is my Sri, so we actually just introduce you guys to what my Shri is now this project is just like network, so spesh, the backend is all pretty much go and the front and like you mean order to make advisor to you know in order to appeal better to the users we have to actually bring in a nice UI framework, like also be using react and material UI for the for the front end. Of course, like you know, you can see it's, you know it's still in the works.

C

You know the UX is getting better day in day out. You know it's much better than like how we presented it out of the previous cube console, so here's actually a dashboard so where we try to present, like you know some connectivity stuff so in in Mussoorie we have this concept of adapters, so you know likely already mentioned like now that we are planning to work with multiple meshes. So here on the left, you can actually see the existing service meshes we work with, and you know here you can see.

C

The network service mesh blow is a bit tiny, but you know, obviously that is one of the meshes like now we were working with. Of course, like you know, that's a very different mesh when compared to all the other ones. There that's definitely something after tail here, so so yeah. So here, so we have a dashboard and you know- and we started off with this deal so the first thing we try to work with is kubernetes. You know now we're working with multiple, but then kubernetes is still our first point of contact.

C

So so I have a kubernetes cluster I've already connected. You know, mesh read to it and- and then here you can see like you know, I'll have a bunch of adapters, but you can see all of them have the same logo, of course, and then for for presenting some some metrics from kubernetes itself. Like you know, we work with Prometheus and Cortana using their API s--. So first things first Marie I think, like an only hasn't actually touched.

C

The part like you know where my Cherie works with multiple meshes, but we also are marching towards a place where you know we want to give the end users the power to actually you know, compare and contrast meshes for their workloads. So you know I mean because we are presented like you know, SEO at so many conferences and people keep coming back to us and asking okay. So what is the overhead of this? Why is this better than link Rd?

C

And you know, I mean all these questions keep coming up so rather than you know, asking them to read some existing benchmarks. Like you know, people would want to actually try out their workloads on the different meshes, so mesh tree actually provides them that capability by you know provisioning this of this meshes and also like you know, loving them to deploy their workloads and then also giving an interface like you know, to perform load tests. So so here is actually a simple interface like Ana, where you can see currently like Ana.

C

We support multiple load generators. How many here, and have heard of w r q2 w RK w RK, all right, yeah, sorry Aeneas! So for Nadia, since we again started off with this deal, I can afford. Io was actually our first go to, plus it's written in those. It was much easier for us to. You know, get started with now. Recently we actually heard about W R K and especially the WR k2 flavor talks, something about coordinated emission.

C

You know, so you guys should go and check it out, check it out, because I'm obviously not going to be the best guy. Who will explain coordinator permission to you guys better so anyway, so so we have to load generators, we support, and then you can see you know by default. We actually collect some metrics from the kubernetes cluster, so you know you can actually see the server-side metrics, like you know, when you're performing a load test and when you perform load test. You also see you know a client-side metric as well.

C

Like you know where you know, you can actually see how the application itself performed from an end users perspective. So that's this page and then the other thing is that we actually provide we process the results in an anonymous way and I love. You know you to actually compare your results, like you know, against the different test runs against different versions of your app against different meshes and so on and so forth. So you know, I can I can just show one like you know.

C

So, for example, so here is actually a graph, like you know, from a recent run, which also where we try to persist some kubernetes information.

C

I know the image is a bit tiny, but if you guys give it a try like and you'll be able to see the stuff, and apart from that lagina, you can actually see the metrics in like a much better way in a textual way like if you're just interested at it, the latency in from and then the same way like you know, we can actually we love comparing results, so you can actually see you know how one version you know behaved or you know performed against the other.

C

So so these are some of the performance side. Capabilities of measure e now before I, actually jump into the management layer like ms or the management capabilities I have to go and, like you know, connect adapters. So here is the section like you know where you know we love you to connect to Pranay DS. So we are I mean like, like I, said, I'm already connected to one, so you know, and it actually shows the context name, the server name, etc. And then this is the section where we will be bringing in the adapters.

C

So I have an NSF adapter, that's actually running on fourteen thousand four. So once I connect again the communication between measure II and the adapters is over G RPC. So that's another like you know popular project, so you can see once it connected like you know, it actually gets the metadata from the adapter and you know you can see the mesh real. Oh sorry, the network, so this mesh logo show up once it's connected like it all on the Left.

C

You can actually see that you know there is a new section in the navigation menu that actually shows up so I'm just going to go there and from here you can actually interact with the adapter to perform some operations on the kubernetes cluster. So now this NSM adapter will actually allow like, for example, to do you know easily provision.

C

You know the mesh of your choice like according to the adapter you choose, and you work with like for examples in this case like it'll, allow you to actually quickly deploy and artwork, so this mesh for you to actually get your hands dirty. Of course, like you know, it'll, we we are starting off with some before the options for now like, but eventually will be. You know, prove creating a more robust UI where people can actually pick and choose and customize like what they want to do with it, and then we have some sample applications.

C

Some of them are already from the networks of this mesh. Examples, like you know, so you know people know about it. So you know, I think, that'll be a great start for you guys to actually get acquainted with along with that we actually created a new app. You know called the hello NSM application I'll be touching on that, like you know, when I do the deployment apart from that, like you know, there was this custom.

C

Config section like you know where you know again, people would want to you know, sink their teeth with llamó, because sometimes it's much easier. So you know we also provide that capability using this custom, config section. So so that's that's the management section right now, like you know, we're mesh. We can work with adaptors and work on kubernetes cluster with the mesh of your choice. So so this is the whole thing before actually going to the next sections. Like you know, Lee will actually continue his talk. Harish.

A

Said it well when he spoke to the genesis of the project, why we created it and why there's a community of people that are helping us create it now part of it was the notion that we well we'd presented that I, don't know at the last, like six cute palms or so and and a couple of those times we're presenting service mesh workshops at the end of which and and a lot of times.

A

I was wrong on sto, but at the end of which people would you know they raised her hand and asked two questions really. This is the same. To that degree should mentioned at first say: well, hey I, know so I kind of get it like the I'm beginning to grok and understand what a service mesh is I'm ready to try it out, which one do you recommend that I try, and so you know the answer.

A

There's disorders generally, like it it depends, and so, in order to facilitate that adoption and overcome the adopters dilemma, we ended up bringing forth measure e that will provision more service meshes than any other project today and I think I. Think the project will probably keep that lead for well forever. I will say the second thing that most of the engineers in the room would say.

A

Well, neither so I see you know again, I see the value I'm ready to try these out, but the way it is wait a second, the I'm getting all these metrics in Piero's I'm, getting all this observability I'm. Getting this security of getting this fine-grained could traffic control over how it is that you know over I'm getting circuit, breaking and re trying all these things that I might have otherwise built into and hard. You know built into my application, I'm pulling that out into the infrastructure and I'm expecting more out of the infrastructure.

A

Well, what's the overhead here? What's the catch and given that each individual project, not all the service projects, produce their own reports about performance, some do and for that they're applauded because performance, engineering and kind of answering the question of how much overhead is.

A

There is one of those befuddling sort of it depends upon what you're running and how frequently you're doing this, and how many, how much you know how much you're asking the surface much to do, even if those individual projects produce their their own metrics they're, not really comparable that you can't really compare across them. It's a bit of apples to oranges, and so that's a bit frustrating so so really measure. He was probably born out of frustration for lack of being with people answer people's questions, so we said off to do that.

A

You know in doing that. You come to. You know that this is a an earlier time in which I consider that well, like any new terms, and as we bring it forth in this technology community, it needs to be defined, and earlier I'd said. Well, maybe maybe certain projects shouldn't be rubbing that serviced mesh word up against themselves, but by that what do I mean what qualify is a project or a product as a service mesh? So so what do you? How do you have to have? What? How do all of these service meshes?

A

Compare architectural or not most of them, look pretty similar. Our beloved network service mesh looks a bit different, which is which is nice and you're, going to think that it's a theme you're gonna, hear about today, most network service meshes and including NSM have what's referred, and for those that, like a Fiat, is, is great by the way, like the more that you're half the audience raised their hand earlier when we said who of you in here is a network engineer or as network savvy.

A

So this is the most the most networking savvy audience we've had. This is great anyway. The point is so these are very familiar terms to you, the data plane. This is the it's a collection of either in SMS case. Potentially, a collection of pipes of connections were proxies, and this is where you lay down this data plane of a proxies or pipes, and they do the heavy lifting they'd. They sit there transparently in line unbeknownst to your application and they're providing network services to be.

A

You know most of your meshes that are going to come with a control plane which is again a familiar, whether you're in networking or not. But this is where you is the operator it go over and you configure you use this as a point of configuration for how it is you want for the mesh to run how you want it to behave.

A

It's also a collection point for policy and in telemetry, and then there's a third layer, and this is particularly nice in this audience- that those of you carrying around your your CC MPs and your CCI ease and etc, and your whatever the Juniper version of that is and another things that you're probably familiar with a management plane and the concept that there's a bit of separation of concern between these things. You're also potentially wondering what's the difference between a control, plane and management plane in context of service mesh well.

A

Measure e is an implementation of a management plan in that it floats above multiple meshes, and it's also going to perform some things that you won't see directly out of the mesh themselves, some of which gireesh spoke to. So, if that's the generic architecture, let's take a look at some of these meshes specifically about their architecture. So we'd said most of them have a control plane and a data plane. This is this Tio's control plane.

A

These are the most prominent sto control, plane, components that we have talks AG nauseam about going through these in detail and Ken what all sto does and how its data plane works. The default side, sidecar proxy sidecar here is envoy.

A

That is a displaced. Both thing, that is a batteries included, but swappable deal. That's not an easy thing to accomplish. Some friends at Citrix have recently recently accomplished this so kind of interesting given the time and how long I like to speak, I'm going to keep keep going and say that hey there are other services. There's a mesh readapt air for octarine octarine is a security centric service mesh. It will lay down either in a complimentary fashion, to sto.

A

If you're running, SEO or not, you can just lay down octarine itself, it does it to uses envoy inside I guess at the little trade mark after the link, your D also here plain data, find different components. Does some different things I would love to talk in depth on each of these, but I'm gonna bite my tongue.

A

Link your D is moving briskly to catch-up in functionality with Sto sto is doing what it can to bring a better user experience and kind of catch up with link or Dean sort of there's a quick tidbit for you. A little measure itself then comes in over top of these as a management plane. It's made up of a few different things.

A

A number of different adapters, we're hoping that the number of adapter that right now, it's kind of a one-to-one adapter per network service, mat or I'm sorry per service mesh type, we're hoping with our advancement of those abstract service mesh specifications that some of that work can begin to subside right now there are five stable adapters. There are two or more in-flight ones that thank. We are three so so in a sec, SEM, Citrix EPX and contain containers continuous mesh and we'll see if the Kong folks step up with cuma anyway. Point is soda.

A

Riche had mentioned earlier and I. Think in our new discussion we said Network surface mesh is a bit of a different breed in an excellent way. So in fact, it's highly complementary to its doing some of the heavier lifting a bit down lower than some of where these meshes focus and thank you and it's kind of interesting actually just by the audience in here that when we said hey, you know, who's networking savvy, who feels comfortable I've got lots of folks raising their hands.

A

That's the absolute inverse and most of the other talks that will give on surface meshes in general. It's mostly developers- and so you can see just by the just in the way in which the project has come forth and and who they're built for different, so NSM can do, is maybe more flexible than my little brain wants it to be.

A

But so thank thanks for that. Listen, I, I get my my confidence and an ego level. Just kind of my wife particularly enjoys it getting sort of knocked back into place. Every time that I go to try to understand innocent. This is you know one version of how it might look. Has you lay down in any on a given node it is. It is doing the things that we talked about about. It does have a data plane, it does have a control plane and a flexible one at that. So this is nice.

A

So a different perspective, that is a Multi multi node.

A

There are also another construct in here: that's not necessary to highlight it, but there's an NSF domain and I get the sense that we hear about some of these constructs a bit later from from through some of the other talks, but pretty interesting system, and if again, going back to my smaller sized brain I like to and and to the fact that as I learned something new I like to do that in context of something else that I feel like I've got a good grasp on and so to me, the closest relationship of an existing technology that I'm familiar with is um is Sdn.

A

I'd done, I've done a lot of Sdn work at Cisco and I'm happy to happy to have spent time there and happy to be off doing other things as well, but there is certainly an SDN sprinklings and kind of conceptual sprinklings here in network service mesh.

A

So with that I think we were going to show maybe a deployment of network.

B

Service mesh there did we already.

B

B

All right so I'm pretty sure everybody is familiar with this commands here. I've.

C

Tried to maximize the size, so you guys can actually see it better.

B

Yeah, that's better.

B

C

So you know I'm just showing that okay I'm just testing if the namespace is ready, so I'm going to actually first show the deployment of NSM in the NSM namespace. So so I have the adapter chosen. You know it's ready to accept, commands and I just gave the NSF namespace in that cluster and I'm gonna. Ask you to actually deploy NSM and here to make it interesting. We can actually.

B

Watch the namespace.

C

Right so you can actually see like in our in in a matter of seconds, it actually provisions the NSM of course, like I, know, like I said we have chosen some defaults and thanks to add, like you know, he actually helped us with a helm last week, but it was like really quick. I did run into some issues, but you know I mean his helm. Chart was like remarkable.

C

I can actually solve a problem very quickly, so so there it is like you know, so we were using helm and you know say: NSM is actually provisioned like in a matter of seconds.

C

Of course, like you know, the other thing, I would say is like I think this is the first live demo of an SM in this cons like we're very proud, the next thing is I mean, like you know, you guys would have actually heard about like in all these other applications, part of NSM, which is a VPP application, the VPP ICMP and then the ICMP itself. These are part of the NSM repo. Now what we actually did was that we we wanted to actually create a new one.

C

Now it has two things like I know: one is we wanted something which we can test using HTTP as a first part like not a second one is like. It also helps us to actually get our. You know get ourselves some better understanding of NSM. So what we did was we actually created an app called the hello, NSF application and I'll talk about that. Like an ass like deploy the application, because you know we were running out of time, so I just clicked it and you can see the app come up.

C

So essentially, there are just two parts: the a and B B is actually the end point the networks of Zen point and a is actually the clients. So you can see it has three pods. Sorry, three containers in a pod, while B has two containers in a pod. Now the other thing is in order to actually make it easier. You know, I also exposed us a you know, create or a cover ad service.

C

For you know a and B and I'll tell you why so here is actually the the port number like on which the service is exposed. So I'm going to try to access it.

B

All right, so, if I then copy it, probably three hundred six seven eight.

C

So so here excellent, so if you guys can actually see the screen, I I did actually so at the end of the part of the URL, like I, actually put a slash k8s. So what we did was we created two two parts to actually showcase. You know the difference. I mean again. This was more educational for us, so here and then I tried to put in a put a diagram which will hopefully be educational.

C

So here when the app a and B pods were actually provisioned, you can actually see that they were connected to two networks. Like you know, one is the kubernetes one, like you know, that's a default and since I did I mean like now the deployment the animal contains the right annotations. Like you know they were given extra network interfaces by networks of this mesh. So the the slash Gators endpoint, like you know, so when I actually hit it, you can see that what it is actually doing is it's talking to so the history.

C

The browser is talking to the service app a and app a is talking to a B over the Carreras kubernetes network, and you can see that's the service name that was used. It's a B, it's awesome! What's that now, I actually did another one like you know, which is the NSM one.

C

So here you can see it is almost very identical image, but, but you can see like what we have tried to show is that the network, sorry, the network, that's used for the communication between a and b is actually the NSF network and at the bottom, like an I've, try to print the NSM IP address, from which you know a god, the response from which was 1061 to so, which is actually an NSA.

C

My IP address, like you, know, arm so so that that's what we try to do, I'm not sure like if we can actually go any much more intelligent than this, but this was like a quick exercise for us to actually, you know, learn and you know get our hands dirty with NSM. You know, and definitely thanks a lot for hit again and again. So so you can see, like you know, would with like matter of like clicks.

C

You know we were able to quickly provision NSM and the sample application and actually see it all in working and then now I'm actually not going to take more time, so I'm gonna try to do it as quickly as possible. I can also go back here.

C

Give it to our thing, and you know the load test will actually begin so so this is again another part of like a measure you like know where you know we perform load test so now I'm, actually using the NSM network to actually perform a load test on the communication between app, a and B, which are like the pods. You can come over like in a while. The test is.

A

And so we just tested here was the the performance. What we just did was generate a whole bunch of load from memory. The management plane slammed that into NS Em's new, well, the hello NSM app using n SMS new network that Ennis NSM provisioned. So you get to that app and then what you can do is continue forth and maybe just based on time.

A

We won't, but you you would, then you know run any other any number of other performance tests, but you then might say well, let's hit the other interface, where we're running just over the kubernetes network and see the performance difference that maybe we can say well hey.

A

This is how much that costs in this environment and weigh that against the value of what NSM is providing my perspective is you will you will be very pleased you can also go through and so yeah, so you can go through and then do some performance analysis of NSM itself, CPU usage right.

C

So this is the overall CPU because we actually provision the sample application and NSM in the same namespace. So you can actually see that the SM VPP forwarder was actually the one which is consuming the most CPU yeah I mean like no conclusions like I mean this is just like our regular analysis.

C

Like you know, because again you know this is part of like what mesh we is doing and what we try to infer from it, etc and then the same way like you know, so we have the memory usage, you know analysis and you can see the VPP folder definitely was actually a major consumer and the NSM manager was also actually like. You know, taking you know, trying to catch up with it. Well all the other ones like no we're. Actually not. You know there were tiny memory, consumers and then oh.

C

This is the the NS es NS he pods CPU usage. Of course, like you know, the site core endpoint did actually like you know, show some increase in CPU I mean like these were actually taken right after we ran some load tests. So obviously you know we expect spikes in a similar way. The NSC had three per three containers in a pod, and you know so. You can actually see how it all performed with respect to the CPU.

A

Context, you quite clearly, unless you have, you know, 8080 vision or whatever. That would be that this is not a zero to 100% scale, or rather this overhead here is quite negligible. The point of the slide, we're talking like you know, is.

B

A

Appointed site is to say relative to of these various components. What do they look like relative to one another as just a point of interest, but at another point of interest, then, is in understanding, NSM and there's a there's, graphs and SM as compared to other network service, meshes of which these slides are something of a teaser. What we have here is a comparison of sto, linker, D and console. Those were the first three adapters that measure II, that that were stable for measure.

A

Ii will go back and do this, probably in an extraordinarily unfair way, for the other service meshes will do this for with network NSM included and we'll have to apologize for everyone else. Looking much worse and but but point is- and this is what part of why we measure was created to help answer. Some of these questions help really help people be comfortable with understanding that, if they're going to get some value out of the system, there's a here's.

A

What that overhead looks like help them tweak and tune things I'm skipping through these in part, because there's a bit of context that needs that that that, if you go to the links that will give, you can read a bit more about the environments these were tested in the type of workloads that these were running, the type of how much load was being generated and it saturates. So the it'll, it's enough to put you to sleep for sure.

A

With that check out. Mestre all feedback is good feedback. Unless you don't like my screen, then you can piss off good good good. Lastly, there's that we're trying to make layer 5 a wealth of information about service meshes in general, if you're inclined come come help us come add to that. You know. Come subscribe.