Cloud Native Computing Foundation Runtime Special Interest Group, 4 Mar 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: CNCF SIG Runtime 2021-03-04

Description

CNCF SIG Runtime 2021-03-04

A

A

A

B

Hey good morning, everyone so just give it a little bit of time and or a few more minutes to see if more people join.

B

B

B

Good morning, hey good morning.

B

B

Are you michael you're, in uh in texas or.

C

No I'm not in texas, I'm stuck in beijing.

C

So it's uh it's a night for me this is my office.

B

Is everything open there? The offices and.

C

Oh yeah in china yeah it's as if.

B

Nothing has happened.

C

Yeah but the uh the following tribals are closed, so that you know um they don't let foreigners come in and are not.

B

What about people from there can they go out.

C

Yeah out, if you god you pretty much, can't come back. That's a problem. If you leave, you can't come back.

B

B

B

I think uh maybe we can get started. I mean this is uh three minutes past they are other people might join in, but yeah we can get started. We're excited to have you here to talk about ssvm, uh yeah, um second state uh webassembly, virtual machine.

C

B

So we know that webassembly is a newer technology and then excited to hear more about it. How you can run workloads using webassembly.

C

All right, hi, hey everyone hi good morning, so shall we wait for 12 minutes more or shall we just start? We can? We can start okay. So um is it possible to give me the um you know? Let me share my screen. You know yeah for sure. Okay, you.

B

C

Let me try that.

A

B

So we usually have this interactive session, so we might be able we're going to interrupt you at times. Yeah.

C

B

C

B

Some questions.

C

Please so um uh you guys can see my screen right, yep, okay, so yeah, it's a first of all. It's uh it's a real real pleasure to um to be able to present to um this group. You know that's uh um so sam I wanna talk about is called ssvm and it's a very unimaginative name. It's called second state virtual machine, also it's ssvm and it's a web assembly virtual machine that is designed for the out of the browser host environment.

C

So, um as you know, the the the web assembly standard was was developed, um for, I would say a second fast. um You know virtual machine long time inside of the browser other than javascript. You know virtual machine, so um it has been do and you know it's. uh It has made some progress in that area, but I think people, especially in recent years. You know coming from the folks from the blockchain world and then from the server side.

C

You know people are increasingly interested in running web at somebody outside of the browser um for full-timers. Like myself, I mean it's, uh it's uh it's very similar to the past that java has taken. You know back in 1997.

A

C

uh You know all the way to you know early 2000s, when I was um you know when I had a startup that doing java application server right. You know um at the time you know we all remember this java applet story right, it's running inside the browser, and then but in the end you know it's, uh um it becomes successful as a server-side technology. That's also where we see you know, uh perhaps uh whether somebody run time is going um time sense change with some significant difference in the back.

C

In the days when we still have moore's law, you know uh performance, isn't such a big issue and the developer productivity is is, is the predominant issue, so the java virtual machine is pretty much designed um to trade.

C

um You know developer productivity for performance, at least at the time when, when when it was first developed right, of course, a lot of optimization has been done over the years and made it much faster now, but the web assembly is uh it's designed from ground up to be very fast and uh to be high performance so that it can, um you know it can take advantage of. You know it can take full advantage of the hardware so, but it also tries to uh preserve developer productivity.

C

In terms of you know, cross-platform compatibility, which is more of an issue now than in the past, because I think we have much more many more. You know operating systems and cheap architectures on the um on the server side than say 1997. When there's you know, um you know pretty much just a few, um you know cpu architectures back then, and now we have like ai chefs gpus and you know a bunch of those stuff, and we also have a much longer history of the linux operating system.

C

We have operating system dating all the back to 10 20 years ago, and it's still being used in production right. So so those are the things um and and- and uh you know that's- I think those are the things um what I think, uh whoever somebody would set out to uh to solve and uh uh we're also where we think that it could be a good fit for the the cloud native foundation. You know, because I have worked with linux foundation in the past, and uh I really like how the uh how the community works.

C

Oh, by the way I used to work at red hat, so that's my uh you know, connections um with linux right, you know so um yeah. So so that's! um But so you know that's the introduction where I come from where's this. uh Where this um you know, ssvm virtual machine come from so it's a popular web and you know well, we call it the popular web assembly, virtual machine optimized for high performance applications and uh um it has a github repository. It's a it's an open source project from day one.

C

So, uh if I may, perhaps I should show you the the github page. You know. That's uh see that. um Can you guys see see this screen? Okay, yep, yep? Okay, great! So you know it's a um for infrastructure project. It's it's fairly popular! It has like 700 stars. So I know it's not it's. um It has been around for two years so from from day one you know there's over close to a thousand commits, but mostly from um you know, from from developers in in a in a closed group.

C

You know that's the second state of the company and people who associated with us. So those so um I wouldn't say it has a very large you know country and commander and developer community. Yet that's also one of the reasons we want to join uh cncf as a as a sandbox project right so that we can, um you know, develop so that we can collaborate more with the developers in the community.

C

So it's a um so um you know there's some people who, like it, you know 700 people who, like it and you know, there's um you know um people contributed code and, uh um as you know, there's fairly comprehensive documentation in terms of where to runs, how to run on different versions of linux, offerings and stuff and how to build it. You know how to um you know how to run for advanced tasks like you know, ai inference, how to um how it works with tensorflow and all that things.

C

So you know that's um so it may be interesting to to um to take a look. You know in terms of you know um what this project is and what's its history- and uh you know um you know, and and and just download the code and play with it and play with it yourself. You know you're very interested, so this is a so this is uh um the project and it's it's a github page and uh perhaps the most important slides um in this. um You know in my you know, prepared presentation.

C

You know just feel free to interrupt me at any time, because um because I want to talk about four uh six application scenarios where ssvm in particular and web assembly in general, um can bring value to the cloud native ecosystem. So you know people say you know we have. You know on the server side we have. You know we have docker, we have kubernetes. We have all those. You know why um you know docker, including kubernetes. Probably you know give you uh the full operating system capacity.

C

You know you can run any application that or any framework that you would normally run on the linux operating system in those containers. Why do you need something um that is potentially even more limiting you know? That's uh because on web assembly you can't run the operating system, so you have, you have a clear boundary in terms of you know what type of application that you can run you you need to conform to some kind of sdk. You know things like that.

C

So you know that's uh um so people ask that question so um so those are the things that we see uh in the market. You know um that's, that's uh people use our product and people use other server side web assembly product to do right. You know, um and the first is what we call gem stack application web applications.

C

So um gem stack. This refers to javascript api markup, you know so it's a the idea is to use a static website generator um like hugo or next yes or um what's the other popular and guess, get gatsby right. You know, so um you use those applications to generate a static website and uh distribute those static web pages through cdns and then um um and then those big front ends have javascript in them. They provide interactivity through apis on the back end, so it's got.

C

It's provided complete separation from the front hand and then backhand, and so the front end can be distributed um any way you want. It can be good on pages. It can be. um You know, um like I said it can be cdns. Can it can even be a blockchain project? You can um say ipfs right, you know, so it can. You can distribute it just like files in any way you want, and then you host the the interactive piece, the web services on the backend.

C

So that's um that way of developing web application has gained a lot of popularities in the past. I think in the past two three years you know: that's, um you know a lot of developers. I love that, especially with uh you know, with um you know, github providing get pages for free. You know a lot of people are doing that you know so.

C

The key issue in jam stack application is to provide simple and easy to use apis on the back end, to talk to the front-end javascript. So whether somebody plays a role here, because it can provide very lightweight uh what we call serverless functions. You know, meaning in the cloud you instead of um the front-end application needs to perform. Some functionality, for instance, send a sms message to to some user for for logging or to create a database.

C

You know something something of that nature that you need a really light function to be executed on the back end, um there's cloud providers who would provide you service functions. You know like aws, lambda and all that, but the problem is um they are fairly heavyweight because you know they are um virtual machine based store container based. So you know so um you basically have setting up operating system and then on python objects.

C

So that's um you know, people have, you know, have a lot of discussions of this type of architecture to say it's, uh it's a it's a fairly wasteful in terms of you know, resource use, and it also has a problem of code start time. It's too too low, and it's also not cdn friendly. You know, so you have all those front-end uh files, static websites and that's generated from your from your framework that you distribute them through cdn for performance reasons to the uh close to your customers.

C

However, your your back-end services is still in centralized cloud. You know people still need to go all the way back to aws. In order to you know, you know log in you know something like that. You know. So that's um that's the ideal architecture. So companies like uh faster, a and uh um cloudflare uh come up with those ideas of service functions on the on edge notes.

C

Right, you know, so they would say you know um we um would provide serverless functions and that execute untrusted user uploaded functions to you know um and that's closer to the edge node on the cdn network and the the the way to do that is not through darker or more heavy containers. It's through very light. um You know language language based virtual machines like web assembly. So that's um that's. um Basically how the I would say the edge computing industry is moving um is trying to move that that direction right.

C

So you know so that's um one category of uh applications. I I'll give you. um You know a full page of links which you can try out. You know the applications we built this way. You know the way. There's ai inference, there's image, processing and all that. So you know those are jam, stack applications all the front-end, it's just github pages and the back-end is serverless functions and the service function is executed on a lightweight vm and ssvm or other web assembly. Vm could fit that build.

C

So that's one area of um you know where we see um you know cloud-based web assembly. Where does it affect it could fit into this? You know jam stack and the second is more of uh more of what I would say. The the traditional jvm or java use cases is to provide a unified api for something that is complex right. So here example, is to provide a unified api for ai inference.

C

So, um as we know, you know, that's uh um deep learning and ai has um you know- has gained a lot of attention in the past few years and uh um to train our ai model. There's lots of books about that, and uh you know they'll tell you how to do python and do that. However, to deploy a model in production has always been quite difficult. You know, there's tensorflow servers, there's you know there are lots of things you need to mess around.

C

You know to get python and everything else work together and it's not super efficient, because um because in order to do that efficiently, you really need access to things like you know and gpu at least but custom hardware is probably the best like aws there's a1 right. You know things like that. So there are a lot of things that you have to do at deployment um for developers to um to write code. That's specific to the developer, to the deployment platform is, I would say, a pre-java days right.

C

You know when you know that's, uh you know you write a web application. You have to know it's underlying architecture, sound spark right. You know not x86 linux right, you know that's, uh so um um I think that's a huge burden for developers and in our particular case you know we uh we partners with tencent cloud. You know they have a. um They have a.

C

They have a serverless business that basically use docker containers to run serverless functions right and the docker containers in question runs uh centos s7, which is about 10 years old and doesn't even run tensorflow in terms of flow. Doesn't even you know you can't even compile for that particular um for that particular environment and uh from the developer point of view, it's it'd be very difficult to get it working because you know um you need to first choose which ai framework gonna work in that environment? Is that epi, tensorflow or oix or something else?

C

And then um you need to um figure out the the the exact configuration the operating system? That's running so that you can static, link, newer versions of glebc into your uh into your stack in order to get running right. So you know we um so the way the web assembly um virtual machine works is that it can abstract away all those issues. So we can have all those issues pre-configured by us um in collaboration with tencent.

C

um You know that's one center for all, and then we provide a language level api in rus and in swift and in typescript right. You know, that's uh two developers and um the api basically allows developer to say you know, load my ai model and here's the input tensor and run the model, and here are the output tensors. You know, so those are the very generic ways to run.

C

um You know um ai models in a unified api and then the webassembly runtime on the deployment platform figures out how to route those those computational tasks right. Could it is this wrong? Is this execu because I'm now running on 10 cent serverless? So this should be wrong. um um You know, should be executed by by tensorflow on cpu or, if I'm on aws. This could be wrong by by say um pytorch by byron x by osx runtime on the a1 process.

C

You know that's so so those are the things that we try to make things easier for developers so that they can provide more another abstraction layer on top of the operating system at the language level, um pretty much things that java has stopped uh 25 years.

B

So I have a question so yes, thank you for the presentation. This is really interesting, so um uh the the web assembly modules when you're serving a model- I mean- I guess you will. You- will still have something like tensorflow serving in the back end right so and.

A

B

And like this will be just the api layer in a in and I suppose uh this could actually live anywhere. It could also leave it at the edge or because they're very light weight modules. And yes, so have you seen some use cases yet where people, maybe where do they put the web assembly modules with respect to the the server and the infants uh where they have?

B

The actual you mentioned that you can have it in aws or you can have it somewhere else, but then um is it better to be closer to the edge or or.

B

You have this really lightweight modules, but then they still need to talk to a backend um and that back-end actually could live anywhere like they could. It could be an aws could be a database or it could be a kafka broker or whatever right does that make sense.

C

Yeah yeah so yeah, so um I think we are still. You know. um The the whole landscape or the whole industry is still in its early stage. You know so um where we want to go, is to have those web assembly modules directly deployed on edge nodes and the edge nodes could be a variety of different nodes. Writings, not just you know the public cloud age where I think are pretty far network agents, but also especially in china. You know you you'd, have you know small data centers in factory where the government has subsidized right.

C

A

C

uh You know that's one of the things where the government has so much power. You know the government tells all the factories that you need to modernize and build a data center, each of them, and then um they have nothing. They build a data center and they have no no, no task to run because there's not so much computational intensive tasks in in a factory right, and so they they unplug the power plug right.

C

You know so so, basically it said they're completely idle or it's not even part of um you know, so, there's all kinds of computational capabilities um out there like that, and uh um our goal is really to uh to work with. um You know uh h cloud providers right, you know to have them because they have access to resources like that they have a resource. The spare computational capacity, in fact guitars for instance.

C

Right or you know, that's uh um you know a lot of those infrastructure projects, so you know um they can all they can put them together and they can run web assembly on them so that they can provide computational services for residents in a city in a nearby city. Right and you know that's, uh um you know all those things are originally designed for the factory, but the factory is not using. It might not use it for something else right. You know it could do um it could process images from the door camera.

C

You know, there's a lot of things that you can do, but you know, um and web assembly now provides a much. um You know much more efficient way of doing that and, as we would say, and as we will see later, it's it also provides. It provides more efficiency on the operating on site. It provides.

C

It is more developer friendly because, um especially for ai applications, because it's uh it's not constrained by the operating system- that's running inside docker, you know, although docker is all the container is, is standard, but you can run any operating system in there. That's become very, very uh not standard right, so you know that's um um so it's developer friendly, it's operational operationally efficient and it's also um um well yeah. That's that's.

C

um You know that's, uh but the trade-off, of course is uh it's not as flexible as docker, so so so right now you see a mixture of those uh boats and uh we would even see uh you know that's. uh um So if you ask me what people are running on the docker, um sorry what people are running on the edge today, you know those are. um You know. That's like qbh projects, which is this cncf project. You know that's uh use kubernetes to to um to spin off.

C

You know, containers on those projects, and you know so so there's a variety of different ways to doing that and uh it's very similar to mainstream. Yet no, but that's we are hoping to make it more mainstream. You know. So that's.

B

Yeah, that makes sense yeah. Thank you.

C

So yeah, that's um so that's really! These lead us to the next. um You know topic which is edge devices. So um that's also where we collaborate with um you know: people like um in the industrial. uh You know situation where you know you have those new software driven cars. You know, that's, you know all those electric cars after tesla they're, all just giant computer sitting on batteries and wheels right, they're, not cars, anymore, they're, just computers, you know, and you need a lot of software in them and uh you know that's people say.

C

Oh, you know the car has a lot of software. That's um and erratically it is you know, that's um you know we worked with um you know, that's um um you know a company that does that they use kubeh and the docker to to uh to run software inside of the car right. You know, that's um because the car needs to run a lot of you know. Third party untrusted softwares use docker to do that, and this is um um for us from our. I think everybody in this understands it's a huge waste of resources.

C

um That's also caused. You know um the same that you heard over and over again the car has more software than the space ship right. You know, that's, uh you know it's because there's so much redundancy so much you know, um uh complexity of running is offering the car, and uh um so you know that's that's also. The area that we really want to improve the ssbm is to make it uh um it's. It is to make it around better on those uh rtos.

C

You know real-time operating system like thread x right, you know, and so that and also access to the um controlling hardware in the car. Through you know what we call the web assembly system. Interface was interface, so that's um to expand the web assembly system in that way, so that it can be, um it can run software more efficiently inside those um inside the cars right.

C

You know so so so that's um the third uh application scenario, whereas the h devices there's lots of different operating system, lots of different hardware, that's um that's that people have to deal with and we we hope weber. Somebody would provide um a better abstraction for for developers to target those platforms better so yeah and then, of course you know the the the fourth one and the fifth one are similar. You know it's uh it's um it's.

C

You have a large system like a sas system or a iot messaging system, and but this when you go to a customer, you need to add um you know functionalities to it. Is that specific to the customer, so the traditional way of doing it is through apis or through um configuration files right. You know that's for easy things, you you try to configure it and then for complex things.

C

You have apis that allow people to call your system from outside and make callbacks, and you know things like that and we think perhaps um another approach where I think it's become more and more popular is to make it serverless so meaning that's in the system. Instead of making callback to external api, I ask developers to upload a piece of software upload a function and I run it inside my infrastructure.

C

That's you know so, and in that scenario, there's a unknown.

A

C

Not uploaded by the developer that has to run inside, you know inside your source system and uh um you need a sandbox for that and the docker is too complex or you know a traditional application container containers too complex and too slow, so web assembly is actually quite uh quite perfect in in this particular scenario, and we've seen this um in quite a few um large-scale projects already. You know, I think in north america, shopify is probably one of the biggest one. That does that you know that's on their application scenario.

C

It's actually also pretty interesting. Is that um you know if you, if you think about shopify, it's a it's. An e-commerce website builder right and one of the heavy customization they need to do is that they need to give people flexibility to provide discounts or rules at checkout. So you buy my shop owner. I want to say if you buy three of those get force for free or buy three of those get the force, get the shipping for free.

C

You know, that's there's many rules like that and uh um in the past they do it with templates, and it's um prop. You know that's address a lot of those needs, but not all of them right.

C

You could argue a better way to is to let people upload a small piece of code, just use software use code to describe what they want, and um but that's creates a situation where you know the the platform has to um specify.

C

um You know have to run this code and uh um you know um in the in a in a shopping in an e-commerce checkout scenario. Of course you cannot have um say a callback, you know something that would take like a second. You know, because you know people can abandon their shopping cart. They have to wait for a second.

C

This has to be done in milliseconds right, you know, so um the the approach they take um you know um is to use web assembly to run this, uh to run the user, updated uploaded software writing so user upload a piece of code that says you know how. How do I provide discount in my in the items in my shopping cart and then you know they use webassembly to run it very fast. Let's check out on their infrastructure instead of the developer, have to set up another server to do the callbacks right.

C

So um this is also the work that we do with um you know. uh Well, those are two chinese characters, but this is basically a tick tock. um You know parent company in china and they have a slack competitor. You could be different a lot you know, so they have a work collaboration, software uh by tick, tock's parent company in china and the um the way that they try to expand their platform is also you know. If you look at um you know, uh their current documentation is uh api, based approach.

C

So if, uh if I, if a user, send a message, you send this message, you know um it. Would you use the api to do a callback to a server? That's the the developer sets up, and then the server comes back with a response right. You know, that's a typical, uh uh uh a very you know, standard way to provide api services, but you know and there's lots of issues with the reliability of the developers system, whether it's inside china's firewall or not. You know where you know there's what type of security it has.

C

You know all kind of stuff, so you know that's, uh um so we thought it's. um You know um an easier way to do. That is to just let the developer upload a piece of code and, and uh you know the platform around it right. You know so so those are the. um So I think those are the two uh related application scenarios that we have encountered for web assembly in the cloud is to have is as an extension mechanism for um for for larger platform. Like you know, a source platform, something like that.

C

So then the last.

B

One sorry I have uh so one question. So can you explain again how how they do that with the with with the chat applications, so they upload a web assembly module to the developers upload a web assembly module to the chat application service and and they provide unique functionality based on that or.

C

Yeah so um say you know, say it's a chat application, let's think about slack right. You know, um on slack platform, you can write spots right, you know the the the boss that can respond to users and the way you you write box is you register callback url with the slack platform to say if my user send a message to the bot.

C

Please forward this message as a http request to my server and my server going to process it and my server can come up back with the response and then you send the response back to the user right. You know, that's the normal way right, you know. That's, um you know the api platform way.

C

We thought a better way is to say: why do you ask the developer to run the server? Why don't you ask the developer to upload a piece of code? The code basically has an interface that says um the input parameter is a string. That's that's is what the user says and the return value is. Another string is what the box says right. You know, so that's that piece of code is running inside the messaging application's own platform. So it's never done and that's meet the security requirements and all that stuff that they may have right.

C

B

Yeah because the same box and all that and yeah okay.

C

Yeah, so so so to to do it in in web assembly. I think the problem is: it's uh is worse in china, you know that's um because in china you have the um you have regulations around content. You know you're not allowed to things that uh to say things that the government doesn't like right.

C

If the developer have their own server, you you have to make some. um You know um to check their stuff, and you know: there's lots of compliance issues and to have it all on your own server on the platform server. You can make it a lot faster. You can make it a lot faster. You know and a lot more compliant.

B

Yeah it's much faster, but also, if you end up with, like thousands and thousands of web assembly modules on your own platform, there might be some challenges there right. So, but yes, yeah because like when you do a callback to an external service, you're doing it to multiple services. Of course you have the network latency issue right and in this case you have the web assembly module and the whole messaging platform, but you may have hundreds of them right so so there may be some scaling issues there, but I think everything has a trade-off right.

C

Exactly but that's that's also why um you know we are interested in working with developers in cncf right. You know, because that's uh because the thing that you have just described how to scale out a thousand wherever somebody instances right, you know how to even start right. You know, that's uh that's something. You know, that's um that's what we consider cloud native. That's um you know, that's um you know, that's what we want to improve on right. You know. So that's.

B

Yeah cool cool.

C

Yeah so then the last one, perhaps you know um it's also a large use case, but perhaps less relevant here is as a wrong time for blockchain smart contract. You know, that's, um I don't know you uh how much um you guys are uh are interested in the blockchain space, but you know um the first generation blockchain is the bitcoin. It's just the ledger right. You know just the keeping record of the accounts and then the second generation of blockchain is what we call easier. You know that's uh it's.

C

uh It has a uh two incomplete virtual machine on the blockchain, so you know, instead of doing coin transactions, you can attach a piece of code with the client transaction, so it executes the you know the the the the piece of code together with your transaction and all the nodes have to come to agreement right. You know that's, uh um however, the the user virtual machine was written by metallica when he's 19 years old. You know it's a it's a brilliant piece of software, but it's not. um You know what we would call.

C

um You know a whale engineer, so you know so there over the years. There has a lot of be a lot of issues. A lot of people lost the money on this thing because you know it's so difficult to work with, and uh um so essentially after inserting you know all the next generation boxing they all decide to choose web assembly, because you know because there's a larger developer ecosystem around it less bugs you know, there's uh lvm compilers and all that stuff. You know, that's, that's.

C

That's completely a complete two chain that you can use. So that's where we, um you know, um you know, that's where we see a lot of financial interest. You know, as you know, bitcoin is like all-time high again, you know. That's, uh so those guys have money. You know, that's that's pretty simple, you know. So it's um um you. You know. So this is also one of the areas we see a lot of. uh We see a lot of interest and also developed interest as well.

C

So so those are the use cases, and uh um so I hope I have provided uh um you know.

C

um You know introduction of how why this is needed on the on the cloud right and why this is not just something you know you should run it inside the browser. You know that's. There are actually some use cases for for this technology on the cloud so.

C

Oh sorry, so the interesting thing about web assembly is that um this has a core specification in w3c and then it has a number of you know um optional proposals and it has a standardization process that is very similar. um Well, I would say very similar, but somewhat similar to the jcp process in the past.

C

So you know people come up with proposals in terms of what you implemented and then you implement that and then the community come up with implementation and then there will be reference implementation and then becomes the the the proposals graduates right. You know so um so, um aside from the the key, um you know virtual machine input. You know specification, that's originally developed for web browsers. You have um perhaps the most important one is the washi it's on on the left side. You know it's called web assembly systems interface.

C

It allows whether somebody to access the operating system features instead of the browser. So inside the browser you don't have the concept for file system. You know, yeah, you have the concept of the dome on the page. You um you, you know so inside of the browser, there's a lot of things that are missing, but if you run web assembly on the directory on the operating system, you need to get access to give it access to sockets to uh to the file system.

C

To you know, that's um so all those things that come standard with uh with operating systems, so so that is washi and the way that the defined washi is expandable. So you can use it to call other native facility because uh what's into cooperating system essentially is to was it for glibc right? You know, so that's uh the standard operating system libraries become available, but then you can have other operating system level. uh Libraries that makes them available as web web summary function. Calls like, uh like I just said, tensorflow right, you know.

C

That's uh it's a it's a it's, not part of the operating system, but it's a native library. So you know you can use it. You can use the same approach to make it available inside the web assembly.

C

So you know so so there are many things like that, and uh um one of the uh very interesting thing um you know um I would say one of the unique features of the ssvm is we try to experiment with as many of them as possible. Try to develop, you know, try to support all of them and uh um and and we because we have many use real world use cases. We also try to improve on that.

C

However, one of the issues with the web assembly community is that right now, there's no. um The w3c has come up.

C

When is the place where you come up with the specs, because there's no place for reference implementations and you know that's a fastly and mozilla um um and I think red hat and the intel, I think, a bunch of them created an organization called the particular alliance uh last year, and I think you know, however, we have reached out to multiple times and other people have reached out to multiple times and it's and it seems that they are still trying to decide.

C

You know the governance structure and all that things you know, so they are not um at least to me. They are not very, you know, very eager to engage with new members or other people in the community, especially with the mozilla layoff last year. You know: that's, uh um you know, fastly, basically hired all the web assembly people from foreign.

C

So now that organization is basically faster, just fastly right, you know. So that's uh um let me see, okay, all right, okay, so so so that's um I think, what's missing from the community and uh that's also where we see um we, we want to be part of the cncf to push um with the linux foundation to push this. um This community forward right, you know to by implementing standards that we can see, we can become the experimental ground for people to experiment, to experiment with more um web assembly.

C

You know um specifications right, so that's um part of the rationale to become the um to join.

B

Another question, so what is it socket is that you said it's under development right.

C

Yeah, so that's another standard to provide network access to the ssv uh different from wesley.

B

C

Yes, so um the standard washi is unfortunate, doesn't include the the network piece. It just has the.

B

C

B

C

Yeah, so it's not that you know that's on this new cpu in infrastructure and all that stuff. So so so that's also our right now to be uh to to uh to try to join the cncf right and then there's um we experiment with a lot of you know um what we call non-standard um web assembly. Extensions right. You know, so we use the wasa-like technique to connect to um other native libraries. So, like I just said you know, that's um and you know we connect to tensorflow and other ai frameworks.

C

We connect to say the blockchain specific stuff like zero under certain blockchain, for instance, you know, there's no concept of the file system or glfc anymore. It's it only has a file account system and coins, and you know things of that nature. So you know so those are the um you know.

C

um The way to um you know owazi is uh is a is a way that web assembly integrated with the host system right, you know, so that's also an area that we really want to experiment with, and that's also, our rationale, trying to be part of the cncf is to become, is to engage to have more engagement with the community so that it can become the reference implementation of what or provide a feedback to spec development at w3c.

C

So that's uh so yeah. So another interesting thing about web assembly is what we call capability based for security. You know that's the concept that has been around for a very long time and uh it's already in system concept is to say you know, that's uh um um you know for each process or you know, for for traditional operating system, um your privilege or your security is being divided by by your user right here. This user has certain amount of um access.

C

You know that's uh um so if I start this process as a root user, it's going to hurt all the access permissions that the realtors can have, but in webassembly it's different you can. When you start the website virtual machine, you can specify the um the access it has right. You know you can say this. It can only access the slash temp directory.

C

So even if the web assembly virtual machine is started by the super user, it can, you can still make it very restricted that only have access to the apis and to the to the resources that you explicitly specify when you start up the virtual machine. So we uh so we think you know that's. uh Perhaps that's some a minor point in in the grand scheme of all the things that I talked about, but I think that's um you know people keep asking is web assembly better. You know provide a better security than docker.

C

You know all the other containers out there. You know I'd say it has the potential to do that. You know because it has a different security model. Does it do it today? Probably not you know it's probably uh because the ecosystem is still young right. It doesn't have time to to fully develop all those features, but it does have the the potential of doing so because it has a different security model. You know so so then this is um also.

C

The point that I touched upon you know is that web assembly is a way for us to um to give um to improve developer productivity when they use. um You know high performance languages like rust and cpu spots, because it abstract away the the differences in the underlying operating system, so it can be um in many ways it can be cross-platform. You know from from very old linux distribution systems in order to wear new ones and also windows and and other things you know.

C

That's um you know all you need to do is to conform to a specific api that rust variety or cpu.

C

So then, of course, there's.

B

Questions that the the ssvm does it have any requirements as to what linux kernel it needs to run on, and there are there plans to support some other operating systems like uh windows or mac os.

C

So so we don't because this is a primary server side story, so um we haven't heard a lot of you know: requirements for windows support, but we can easily do so. It's uh it's another reason to join the csf right. You know this is to have more developers working on it with different needs right and the the thing that is high on the roadmap is to support the rtos x-ray x. You know so that we can well, you know, very limited. You know um h computing cases like in automobile right.

C

You know, so that's uh so! um That's where we think this project is ready to be part of the community is that you know there are certainly a lot of things that we prioritize, because our business needs right. You know, but you know we, but we can also see there are other people would want to. You know want to add other things to it. You know so we love to see people have a wasting implementation for windows. For instance, you know that will allow it to be yeah before there are windows that makes.

B

Sense makes sense yeah. Thank you.

C

Yeah, so the cloud native support.

C

The thing I want to talk about is you know, so those are the use cases that we currently have in the cloud, and one of the things that I want to emphasize is that we are, um you know we are in the process of making uh the ssvm web assembly virtual machine fully compatible with kubernetes by conforming to the openstack open container initiative. So you know so that allows us to use um you know cubase and all the kubernetes compliance on tools to to start and manage. um You know web assembly runtimes.

C

So that's um you know uh partly answers your question. You already questions that, if, if this is getting better to a large saucepan or you could have a southern instance, it's how do you manage them right? You know that's uh so, um hopefully that's um you know um that in the near future we would be able to use. um You know the kubernetes to be able to do that. So.

B

One question is: uh have you looked at this project called crosslit they presented? uh Yes,.

A

B

They're, pretty they're they're they're, pretty experimental right, but I mean their idea is also to uh have web assembly modules run with kubernetes. So it's just something like it's. It's gonna be similar to that or it's going to be slightly different or or yeah.

C

I think the goal would be pretty much the same, but the technical approach might be a little different. You know, that's. We modify different parts of the operating system in know to achieve that, because you know right now. It's all um you know tutorials. You have to hack the existing system.

C

In order to do that right, you know yeah, but that's uh um you know we just hack at different points, but that's uh um but but I think but again I I think those those are the things that we're going to experiment and and and a clearly better solution going to emerge, and we can all use that you know so yeah yeah, coco, so yeah another thing about um um using weber, somebody in in particular you know compared with java the jvm like I I draw a lot of you know: parallels between website web assembly, runtime manager and the jvm.

C

So it's uh it truly support multiple languages on the front end. So um I I often say java plus jvm, is roughly equal to rust, plus webassembly. Today you know, because rust is probably the best spotted front-end language for web assembly, but it's not a longer one. It's uh swift is well supported as well, see us as well supported c and assembly scripts and all uh and those other two are. You know uh blockchain languages like fe and solidity.

C

Those are you know, programming languages that come on the blockchain, so those are well supported on webassembly. So it's it's different than java, where you know there's multiple languages on the jvm or jvm languages. You know all java, you know like languages right, you know so so those are the difference, and uh um so this is pretty much.

C

The last um you know important slides is to say um we did a lot of optimization for webassembly for for for representative two years to do that, and one of the most significant things we did is aot, you know, is to do ahead of the time. Compiler optimization. You know um this. um You know when web assembly is used inside the browser, um the the the way to optimize it is to use jit because inside the browser you never know what what code you're gonna run into next.

C

So you know so you just have to optimize that as you compile, but on the server you pretty much know that you're gonna run this piece of code for some for a million times two million times right. You know, so you can afford to do ahead of time. Comparing you know, that's the first time you see this go become hard, you know that's, then they run much faster.

C

So that's um so so we so we wrote a paper and published that on triple is software earlier earlier this year, and this was published in the january issue and uh I compared um um uh ssvm performance with the other leading web assembly. Runtimes, like was v8.

C

Like I mean um you know um fast days through set and uh and also compared with um you know normal docker, it's darker. Without anything else, it's just a plain darker, plus c, plus plus application. You know there's no node.js or python or anything like that. So it's fastest way for you to run docker, you know. So um we did all those and uh it seems um you know um that's.

C

You can read that paper and- and I think you know this is vm rom- um you know- can outperforms most competitors by very large margin and I think the the reason is aot. You know that's uh um in some cases we, um you know the ssvm runs faster than native. You know without docker native and the um the reviewers was very surprised by this result and and and and they sent the article back to us to say you know this must be wrong, but uh we run the test.

C

It was right, you know, because um with aot you can optimize for the machine you are running out. You know that's uh with the native. You are just optimizing for a class of cpus right. You you that's uh so so um so yeah. That's um that really showcases the the the performance characteristics of you know.

A

C

um You know the the things wherever somebody can achieve so um yeah if you're interested- I I would send this. um You know uh slide that to to the group that there's a lot of examples that we have done. You know, that's uh you know, do the image processing add a watermark to the image flip the image to do the ocrs. You know, that's. You have an image that I can read about text on it to do. um Tensorflow. um You know image processing like all kinds of image, class classifications. Those are all live.

C

Demo um links, you know, they're live because you know it's uh because of thanks to jam stack right. You know on the on the back end, it's serverless. So if no one use it, we pay nothing on the front. End is scatter pages, it's all static website web pages, so it costs us nothing to keep this demo live. You know, other. uh You know. um uh Unlike the old ways of doing web application, I have to have a server. That's that's that runs all those demos. I actually have no server.

C

You know it's uh so nobody use it. I pay nothing, you know! So that's that's! uh That's! um We just thought: that's, that's a very where um you know very good way to to develop software. You know so then there's tutorials, we we write up very comprehensive articles in terms of how you know to walk through the code, how to develop those applications, and you know things like that. So yeah, that's it! You know. That's uh um you know um sorry, I'm long-winded at some. Sometimes, but that's that's my baby right.

C

You know, that's my you know, that's uh that's my project, so yeah, that's um yeah. So if you have uh yeah feel free to ask me any question.

B

Yeah, this is great, I mean I think it's uh it. It will be great for the cncf. So you are you thinking about uh applying to sandbox or for sandbox, and did you do you know the process or or yeah.

C

So so so I think you know um um we are um figuring out the forms. I I think that the next review date is a letter this month right, so you know, so we still have time for the for the next date right. Okay, that's! I.

B

Think I think so yeah yeah there's, uh I think, there's a form that you have to fill out and then it gets considered by the toc and and then typically, if you have all the requirements like the readme, um and I think uh you agree to the transfer of the assets and all the type of stuff that you do with open source. um Then it goes through and it gets accepted into sandbox. So, okay, so yeah, and I think it would be great. I mean the the currently you there.

B

There are no uh projects uh runtime for webassembly in the cncf, so this uh I think, we'll get more exposure.

B

um There there's also some events at kubecon, there's um cloud native webassembly that I think it's called so there's going to be more more um presentations about the topic and I'm not sure if uh this the cfp is available. But it I mean something I'll keep in mind if you, if you want to present yeah we'd.

C

Love to yeah, that's yeah, that's um we definitely would uh uh apply for for the sandbox. I think for consideration. For for the next meeting, you know yeah and uh you know to to participate in more sincere. You know activities, that's that's! Our goal is to you know, engage the community yeah yeah.

B

Great and um yeah you need also you can share these lights on on the secret time, slack channel. So that's maybe some people might be interested in and maybe they have questions or whatnot or maybe they can. They can uh actually contact you if directly.

C

Excellent yeah, so I put my contact information at the top of slides and you know, and and and share it yeah. That's that's! That's a great idea.

B

All right well, thank you very much and yeah and then we'll we'll keep in touch yeah. So thank you yeah and hopefully you can. You can travel soon too.

C

Yeah and uh hopefully we can all travel soon exactly. Okay, thank.

C

C