.Net Foundation .Net Community Standups, 22 Oct 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: ASP.NET Community Standup - Oct. 22nd, 2019 - HTTP/3 and QUIC with Justin Kotalik and Andrew Nurse

Description

Join members from the ASP.NET teams for our community standup covering great community contributions for ASP.NET, ASP.NET Core, and more.

Community Links: https://www.theurlist.com/aspnet-standup-2019-10-22

A

I've decided, yes, Wow I am excited about today's show, as I told you before, I love, the new shiny. You know maybe breaking demo bits. I'm gonna fall.

B

Fun I think justice told me Justin told me on the way over is like there's a bug in my code. I really hope I don't yet it's like. Oh yeah object disposed accept or no the object.

C

Was already garbage collect? I'm like oh, you know what it's prototype for reason: yeah, okay,.

A

So so both of you have been on the show, but it's been a while yeah.

C

A

You introduce yourselves.

B

Yeah yeah I'm a understand nurse I'm, an engineering manager on asp.net core I've, been on a couple times. Talk about signal are some other things like that: yeah yeah and.

C

My name is Justin I've been on the astronaut core team for around two years now, two years and some change and I primarily work on like guest role is an CM, and now something should be three stuff, which is pretty awesome. You were on the stand up once before. I was on the stand up three years ago, when I was an intern. I talked about URL rewrite the middleware, I wrote as an intern and hey we shipped it. So it's not gonna. Be that bad.

C

A

Know it's great it's one of those things where we you need it to work well and and I've had it save the day. For me several times, yeah like.

C

Once I remember, hearing was I. Think Hansen was talking about how he had a URL rewrite file as part of is, and he was like I'm pouring my cycle. Linux, like I, can't use these year. I'll rewrite roles with ice at all. I just brought the file in called you like use your own use, rewriter and pass the violin, but the is rules and Ashley just all worked, which was. It was a good cyril. 'if it was like. Oh yes, we actually rather beat you that, like worked bugs bugs, was awesome. Yep.

A

Alright, let's jump right over to our community links for the week all right, let's find my screen we're set here.

A

Awesome I'm starting right off the bat with a Turkish post, I, hope, you're ready for reading some Turkish or I'll translate to English needed. So we are trying to do this every week. We want a feature at least one non-english post and unfortunately, there's a ton of great content out there. So Emory is writing about content security policy for dotnet, core MVC projects and so there's different ways to handle this. There are some some libraries out there that will handle this.

A

For you, there's 1, MW, SEC, there's a I think we featured a post from Anthony Chu a while ago. People have written tag, helpers Tomas did something on that a while ago, but this this is this, is you know a nice walk through? This is writing the code kind of from scratch. It's explaining why you need content security policy and so I always feel dumb when I say this is I, think it's nonce yeah all right! We're agreed it's not so you so you need. This is basically like a hash right. This is a generated.

A

It's a one-time value and so an attacker is not going to be able to create that value each time. So here he goes through and uses the orangey crypto service provider and shows writing that out explaining you know what what is valuable for, and so that's that's pretty much it and then showing. If you, if you don't have you'll you'll, get an error, which is what you want to see so nice.

A

Alright. So here is a repeat: offender on the community, stand up and relax he's going through a series on dotnet core 3, and this is writing about local tools. So we've had global tools for a while, and a niche been to with global tools is if operating with someone at a you know, PayPal where you're sharing code and your project is relying on a global tool. Then you kind of you need to tell people hey. You need to go, get this tool or whatever. So with that an acquire 3, we have support for local tools.

A

That includes a creative tool manifest and going through. You can install so here he shows in and cake and then for your for people that join. You know that are using your project. Then then they will I may have scrolled by it too quickly. You you can. Basically you can say here: it is here's the exciting part, so dotnet store and that'll political tools. You can just run if you had your own version, or this will show you. This allows support for specific versions.

A

So that's you know pretty straightforward, I, think and and it's nice to be able to have that so that you can take an investment in global tool from within a specific project. Well,.

B

Yeah and I think that also allows you to do things like have different versions in different folders, which is probably useful.

A

Because, like when, you've got a global.

B

Tool, that's doing something like say generating code or it's like a key part of your build system. There wasn't really a way to a make that so that when you restored, they just appear and be managed the fact that well what if you're switching branches, what, if you've been switching projects all the time so yeah cool.

A

Okay, yep about the versioning that.

C

Which tool wins first I.

B

Think the local wins first.

C

B

Local tool installed and you've also got it installed globally. I think it pepper.

C

Is the local that makes sense yeah.

A

Alright, just a quote there from the identity server folks, so they Dominic announcing that the three oh version is out: support for dotnet re SDK and some other nice new features. So that is cool. We are yeah. Alright. This is one that I thought was pretty cool. So on kit it is Oktoberfest and I got a chat asking you know, hey or actually got a pull request, updating the dotnet foundation website. We were previously on dotnet core to one and so I got a pull request from him.

A

Saying: hey, I, updated you two to two and me being you know, always wanting more I said great, how about three? So so, but actually I'm kit went through and not only did you know sent me a great poll. Qwest I was able to you, know, merge it and they were just great I'm. So this live in production now, but also he went through and wrote a blog post explaining. You know step-by-step what he did all the changes along the way.

A

So I've done this recently I did on a Twitter stream last week, where I updated a site to.net course or asp, net core 3 and just going through the docs I felt like the docs were pretty straightforward. So this this felt pretty familiar to me, but so just uh the steps on packages, the change from my hosting reference I web host and I, boosting environment, I, host web host environment, yeah I can't say that more than once fast.

B

A

Yeah yeah and then, of course, changing endpoint routing so and then and then also getting rid of the how the compatibility version 4 to 2. Now question on this I remember when a compatibility version came out, and that was a thing that allowed for you know like some things that would have been breaking changes and allows you to kind of select your behavior. There were none of those that really applied in dotnet core 3. Is that kind of a as neat like this compatible version thing or any time? Well,.

B

I expect I'm gonna speak relatively under informed about this, because I'm not entirely familiar with it, but I believe it's one of those things. What we that we do between major versions more to allow you to opt in and out of new functionality added in toilet, basically, there's a lot of stuff going on in the guts of MVC that we want to adjust, and we we do that in ways that we believe are non-breaking.

B

But sometimes you know some edge case comes up, and so the that's designed as a as a way to let you say: okay, I upgraded to 2, but my site needs me to roll back to the one behavior for these NBC specific settings. I, don't think we do it in three. Oh the ideas in three oh you're, sort of reacting on the the current major version, but I would double check with someone like Damien or something I'm, not a hundred percent on that I'm.

C

Not certain four three one: if we're gonna do it is yeah.

B

So just kills yeah. So there's my completely non authoritative answer: I.

C

Would also say like at this point: you should be upgrading to 300 and so upgrading to two because well.

B

Yeah two is gonna, be end-of-life and three one soon so because that's gonna be the LTS version, because 300 is not a long term support, but three one is gonna, be the price.

C

Period, the upgrade from three out of three one should be pretty mild. Nothing yeah, hopefully not if it's more than something that yeah now.

B

It'll be a pretty smooth transition, we're looking at there cuz three one is really about those kind of refinements to make it ready for three years of long-term support.

A

Alright, so this is nice as from Bradley he's using sequel mr. mcpeet uh ba bill msbuild asked and/or did so just like walk through this is this is doing the PWA magic, creating a manifest jason and service. So he walks through he's, got a calculator application setting it up with the things you know like icons and cash and actually pretty straightforward. There's not there's not a huge amount of code in here. You know some general kind of styling and stuff, but pretty quick walk through so this.

A

This is nice to see there yeah and you know something like blazer makes sense. You know I think in a lot of cases to do a waa, so yeah alright. So this is, let me see: okay, so Chang, we hopefully I'm saying that kind of right. This is interesting. I saw a discussion on Heroku today, and this is neat on using github actions to deploy asp net core to Heroku, so it's you're able to containerize- and you know that works really well with github actions so kind of over there.

A

So here not at all a Roku expert, but this was relatively easy to follow. So you know going through setting up all the the docker izing, so the docker build then going through and creating the Heroku app. This was the part that was kind of new to me. I haven't looked at Roku for a bit, so you know creating the Heroku application and then just you know, building that in through the workflow that github workflow to actually deploy that out.

A

So yeah, pretty cool and I said, have as an option yeah alright last link for me to share. So this is exciting: ABP announcing a 100 release, so the this is a BP this. This has been a long time building to this release, as they say here, three years of continuous deployment, so 5900 commits so there's a lot.

A

This ABP project, we've featured them in the past before there's a lot of stuff for all kinds of stuff, modularity multi-tenancy, you know so a real kind of a boilerplate sort of application and so quite a bit to dig through. So congratulations to the team for the big me and that is it for my links and I will be sharing these out in the all the places on the the notes and the in the chats and yep I I am about done. Talking. I am excited to see some cool siding gone.

A

Do you, you know what a ABP.

C

Stands for so yeah.

A

I think I thought it originally stood for asp net boilerplate, but I think they have rebranded it over time, and so I think they've gone to just ABP. I think that there was more than one project that had a similar kind of name and so that I think that was part of it, and so.

B

Yeah, it's kind of just thing that was like acronym names that then they just decided to drop the acronym but key, the name like, for example, and or they rebrand a little bit. I know gr PC was Google RPC, but now it gr PC stands for G RPC remote procedure call. So the G&G RPC stands for G RPC, it's the classic recursive algorithm, yeah yeah another.

B

Another interesting version of that is the quick UDP internet connections or qu IC, which is quic, which is a new transport protocol that we're going to talk about very soon which the the recent specs they change it to say, quick is just a name. It's just a name doesn't stand for anything anymore, but I think originally it stood for something like that. Quick UDP, internet connections and on.

A

B

A

Well, so so I I did get in the chat. So Riyadh is saying that a BP, it's a complete rebuild of asp.net boilerplate cool.

B

Yeah but yeah, so three go Donna core 300 shipped pretty recently and in that one of the big pieces of news HTTP to full support on by default, but we are not resting there. So today, we're gonna talk a little bit about some prototype work. We're doing at and and the plan is to ship this in a future release.

B

We don't know exactly when yet, but we're looking into H 2 B 3, which is the next version of HTTP, turns out there's already an HDTV 3 out there and HB 3 is sort of a new version of HTTP built on top of quick, that's qu IC, and let me actually hearing these throw over to my screen. A little bit quick is is the the quick website. Quick is a new sort of low-level transport protocol, so think quick replaces TCP at that level.

B

So it's a way of sending reliable messages back and forth and then on top of quick you can run HTTP and that's what they call HTTP, 3 and I.

B

Think Justin's got a little bit of content to go through to talk a little bit about where we're going, but basically what this comes to is we want to get out and get started running with HTTP 3 and start playing around with it and start sharing it, because we don't know exactly how much work there is we're still sort of figuring that out, but but what we do know is that people are starting to use it CloudFlare just announced recently that they were supporting HB 3 everywhere. So we want to sort of share.

C

B

You what we've been working on yeah.

C

For sure yeah many other companies, Google Facebook of all than kind of looking into quick support, so gonna give again I'm Justin I'm gonna give a very brief overview of what we're gonna be, what our current plans are with dotnet and then htv-3 and quick support so yeah.

C

So first, pretty clear question is why the heck do we need another version of HTTP, so we implement HB, that was, the spec was pretty much done around 2015-2016 and it introduced many many cool things over H, b11, I, introduced multiplexing being able to you know, have multiple streams and connections being able to have multiple streams go at the same time the representation be requests and use flow control and header compression as well, in order to just speed up the protocol itself. But fundamentally, HB 2 has a few problems.

C

The the first one is really that when you are doing the streaming connection, these are all still happening at the application layer. So what happens is when you get something called head-of-line blocking in with the TCP layer? So what head-of-line blocking is, if you have like, let's say, five people waiting in line and the first one is taking forever all four, but behind it are blocked effectively.

C

So an HTTP one, you can have multiple requests going on the same connection and if the first one I was taking a while all the ones behind it were blocked right.

B

Cuz in HP one, you had one TCP connection and you sent one HTTP message and you had to wait. It had to process that before it could process the next one exactly, but in HP you're still on that one TCP connection, but now you can send a bunch of different parallel requests exactly so up on the application layer. You have a bunch of parallelism, but.

C

A problem is that still a TCP layer, if you lose a packet for whatever reason, all the other packets cannot be processed because TCP is reliable and in order transport. So you have to wait in that.

C

It's kind of a fundamental problem and htv-3 does address this by using UDP and with quick right. So unless quickly, I'm gonna hate myself. If I say too many puns quickly,.

A

Go over UDP already quick, so.

C

First, off quick is a UDP based protocol. um It is I, have a little diagram on the top right effectively. What we have is htv-2 you had this different network stack with TLS and TCP, so, for example, and kestrel. This would look like you have a socket and you're reading TCP from it. These TCP packets eventually go through SSL stream and then after SSL stream, you're, eventually in kestrel and you're, parsing and processing these frames with htv3.

C

This changes a lot all of that stuff with TLS and everything like that is now inside the quick layer so effectively, you're gonna be reading from a socket you're gonna be getting UDP packets and then all the stuff. That normally would happen. It's all processed by quick and.

B

I saw someone post a comment on the chat that oh UDP isn't that unreliable yeah right and so that's where quick comes in so.

C

Quick effectively implements reliable transport data transfer and also in order within a given stream so effectively reemployment CCP, which is a lot of work, and that's the reason why again, not many people have to try to implement quick, it's a very hefty spec, to say at least we can show the spec later, but it's just massive yeah.

B

And I think the thing that's different there about what TCP does is where TCP does reliability as well, messages come in. If you get a message out of order, it blocks and waits for the message to come in they're all like there's retransmission. The client will keep resending if you haven't acknowledged things like that, but it's all on that one big connection pipe.

B

What quick does is reimplemented that, but instead of one big connection pipe, there are an practically infinite number of streams and you can just create new streams as you need to so, even if you lose one packet for.

C

A given stream, all other streams, can still be processed at the same time. That's the big thing so friend. If you do lose a packet for a given stream, that's that's. Individual stream will be blocked, but all the other ones will allow it to be well will be able to process same time yeah. It's also very analogous to what h-2b to did with the connections and streams. So HTTP you have these, like. You have an HP connection and you have streams on top of that effectively.

C

The same thing happens: it's quick layer, you have a quick connection and then you have quick streams on top of that as well. Quick also introduces some ideas of quicker round-trip times, so normally, when you are negotiating or setting up a connection, you have a lot of like handshaking and a lot of back-and-forth between the client and server. Quick makes a couple things really quick with.

A

C

There's a lot of things to improve this experience. um I I'm not quite sure out what happens at the protocol area. Yeah.

B

I mean I I have not read that protocol or spec in-depth either, but yeah. It's about reducing those round-trip times pretty dish, because because in TCP and with TLS and all these you've got, you got to do a TCP handshake and you've got to do TLS negotiation to figure out like okay. What certificate are we using? What keys are we using quick, does a lot to try and roll those all into one set of handshakes and also reduce the time it takes yeah and.

C

So with that quick one of the biggest appeals of it is the amount of time it takes to set up a secure connection. I remember seeing some data about it being a few orders of magnitude faster than sanic the normal TCP connection, which is nice. um Another thing about quick kind of implied by that is that quick, it's always secure.

C

Tell us one point: three is required by the spec I'm, not sure if technically TLS 1.3 is required, but I think there's a lot of stuff in TLS 1.3 that improves the connection negotiation, speed again I saw somewhere. It may not be required, but like by the spec Israel yeah.

B

You have to use at least TLS 1.3 and they reserve the right to change it in future versions that were saying yeah.

C

That's just a general overview, quick! Well all right now, once you have quick, you can implement H 2, B, 3 and effectively it's just a brand new way of sending HTPC HTTP over the wire as a binary protocol, just like HTTP 2, but there's some small differences between how the frame types and protocol and all that stuff that's fairly minor, but there's a lot of similarities between htv3 and HTTP 2, but it also has some HTTP 1 elements in there. It's just an interesting mix.

C

Now that streams and connections are handled by the transport, you don't have to worry about flow control or we're handling stream IDs, and they are anything in frames. It becomes a lot easier to implement at the application layer. So basically.

B

They got to remove all that stuff from hb2.

C

B

Quick has that yeah.

C

Exactly so ends up being a lot more like you have a single request and you don't need to worry about a bunch of like multiplexing at all, because all of us handle that the TR at the quick layer, which is nice one other thing that was really nice about htv-2, is something called each pack effectively has header compression allowed you to have like references to a static and dynamic table.

C

It allowed you to compress your headers and really small values for common headers and also like for if you have like a request, and you want to be able to save headers on the server to be used with an other requests, you can do that as well. It's.

B

Like if you have to send the content-type header you're, not sending those characters, every time you spend sending like 42 yeah, it's just a header.

C

B

C

So, Anza being really efficient with that yeah, there was a couple reasons why they needed to make a new one called queue pack, instead of H package, the kid couldn't reuse it I believe fundamentally, is because with HT be to all of the headers for coming on the same connection, and so I think they didn't need to worry about like concurrent headers, because now the headers are being sent on streams that I think had to make a change to the product. Yeah.

B

Hb2 made a little a few concessions to the fact that, like it allows you to interleave things so that what you can have data coming in from one big giant upload, while other requests are going on, for example, but it made some limitations. That's certain things have to happen completely before anyone, any other requests can be processed and because quic is all about splitting that down two streams, letting you have a lot of different parallel streams. It's multi-threaded, TCP I, think is one of the comments I saw in the chat.

B

It's like it's an interesting little little way of looking at it cute. They had to rewrite that that logic and change it and that's what coop XO present.

C

It's very similar to a CH pack with just some changes and because they're chain making changes, they made a ton of good changes to the static table. I remember the static table needs to be to only had like 30 entries or something like that, and now they have like 100, which is really nice, because there's a lot of common headers out there and also just as I think.

A

About it, Internet traffic that goes on and we're constantly sending the same string over an hour ridiculous. You know yeah.

C

Yeah yeah for sure, um just a quick shout out to a website I've been using it's I John. Is there gonna be a way to for me to share a bunch of links after this potentially yeah.

A

Send them to me and I'll add them into the community awesome.

C

That'd be great, so this is a nice community link. I wrote this, but um it's pretty much gives you a quick, cannot use. The word quick gives a nice an explanation of htv3. Why they're wanting to do it over HTTP to what problems that has? What is the yeah? What's the the benefits of it and a bunch of other things about it? So just a quick shout out to this.

C

Alright, also mentioning that, so you can do HTTP 3 over quick, but quick is effectively just a way to transport bytes between a client and server and vice versa. So there's been a lot of talk about doing non-http workloads over quick, one I know has been talked about, is G RPC I know it's been kind of briefly mentioned on the repo or the GRP cero repo I'm on it wasn't certain if that's gonna, be something that's happening sooner than later.

C

There's a lot of questions about if the quick, the current quick spec is actually viable to use for non-http workloads. I know that they're talking about a quick v2, so that also made me something. But another thing we try to play around with the signal are and I'll eventually show a demo of a single are using quick, which is yeah.

B

So this is signal are running directly on quick with no HTTP yeah. Exactly and.

C

Then finally, DNS there was some talk about DNS, I, don't know, looks, look up, is UDP or not, I'm, not sure yeah I think it is I, think so I think so anyway. So we're talking about doing I, don't know something right, quick with DNS and doing that instead, alright um and then yeah some shadows to libraries that are currently available right now. There are many many different, quick libraries today.

C

The first one is and as quick is one that's being developed by Microsoft is written in C and it's actually the one we're currently prototype prototyping with we'll go into more detail about, like some of that stuff, we're doing with them as quick a little bit later. Other ones include quiche wishes and our pun on the word. Well, it's fun and quick and something that it's about. My CloudFlare CloudFlare announced that they are using htv-3 on both the client and server I.

C

Believe um Google, quick, I, don't really know what the name of it is I like the Facebook one though yeah I was I. Looked at that and I was like what is most yeah. It's me: it's you're supposed to say, has move fast and then they have. They have a quick server as well neck CO neck, oh I, think that's Mozilla version. um It's also written in rust and there's again a ton of others. So I'll show you a quick interrupt matrix. It is so.

B

Wait a lot I think this is what this matrix is showing, and this is an illustrative thing for why we're talking about this is an experimental early, prototyping kind of phase each column and row is an implementation and the boxes are like how they work together, so they don't even all work together to the same level. That's it's is. This is a constantly evolving spec right now, yeah, it's not even in felt like the spec itself is constant, is still in draft right now. Yeah.

C

B

C

To settle down a good amount, but there's still problems, Irene discussed and I think they may wait for v2 at this point. But a lot of a lot of question marks, though, about the actual, like quality of implementation, there's still a lot of Nick's or kinks that may be worked out still.

B

Early days and a lot of this.

C

B

C

Exactly so just an example is we have the kreb project is chromium, slash, quick. This is what chrome uses for htv-3, quick support and.

B

I think this is another thing it's like, so chrome does have support yeah for El version of quick. Today, yeah and I think it's one specifically google's using on their servers and then they because.

C

B

Really involved in developing this yeah, but one of the things we'll show and actually I think we could probably jump towards that as we can show a demo of what we've got in an initial prototype, using chrome to actually make an HTTP 3 request to an eight-speed on our servers yeah. So if you wanted to show that now, let's.

C

Start with that, do that now.

B

Yeah we're just doing and then we'll talk a little bit about what.

C

We're gonna do sure what our thinking.

B

C

B

Could show the startup and then we'll have you zoom in a bit when you find the file, you want no all right. Let's zoom! In a little bit on the text, there yeah look.

C

B

C

It looks just follow me. Thank gosh yeah, all right, so this is your normal program. Man. The only difference is that I'm calling an API, it's called like use, ms quick, so what's doing is this is effectively adding just like we had. We have different transports in Kestrel effectively, you can call you sockets use Livia V. This is just another transport, that's implemented with this bedrock API I'm, allowing you to get a connection and get a stream out of that and stuff like that.

C

So what we're doing is we're adding them as quick as our ability to get a trace for transport to send and receive bikes from after that, we're also listening on both HP 1 and H 2 P 2, as well as HP 3 on this one I a little bit hard coded right now, but should be ok so now, if we control f5 and run this sample.

C

Hopefully, hopefully, when I refresh this page in this builds, we should be able to see HP 3 working, even though I definitely have a look that I need to fix, but we won't.

B

We won't talk about the bug; hopefully it won't hit the bug, hopefully on it. Ok, if you keep talking.

C

B

C

Running a computer, it didn't happen at all, oh so we're listening now. One.

B

Of the things yeah, why are there two URLs there with this? Is it says the same port number right.

C

No, it's really interesting, so you listening on both TCP and UDP, are effectively account is different. Yeah.

B

They're different port spaces, like port 80 on TCP and point 80 on UDP, are different. Uh-Huh.

C

So a lot of times when this is specifically from I can tell is when they have quick input on both on the client and server when they make a quick connection, they usually do it to the same for for 3 port that you normally would like.

A

C

So, for example,.

A

C

Could go to youtube.com. You would sometimes see like the same request going to port 443 buzz using quick instead. So now so.

B

Now, because you're listening, you're doing HTTP 3 and we're doing it, we still listening with HP 2. That's why you see those two now listening, our URLs haven't been updated, but it's saying I'm listening on TCP five, five, five, five and UDP Wow.

C

So different I know it's really interesting. um So if we go to here and we refresh, we should be able to see you're.

B

Not showing it, you got a right click right, click on the column and I think what.

C

Type protocol, h2 plus quick 99, is actually htv-3, so yeah google chrome reports. They should be three as this for whatever reason I don't really know: um HTTP 2, + click.

A

Quit 9 quick.

C

Quick 43 I think was the Google's implementation yeah.

B

C

Be three slash, quick, it's kind of weird yeah.

B

So like- and this is exactly it- it doesn't say- HB 3, because even that's kind of a new idea calling at HB 3 for a while. It was just called HTTP on quick and then they.

A

Kind of decided.

B

Let's just call it h, p3 and let's just make it the next version yep so um do. Let me take a wire short trace of this yeah.

C

That's good yeah, but we've.

B

Got a bunch of so um one of the things I think we should talk about is okay. So what does this mean for net core? Because.

A

B

A great demo it's been hacked together, like what does this actually mean and I? Think there's a the thing to remember about supporting something like HT p3 is there's a whole bunch of elements to it right, there's getting it working in asp net core on on kestrel there's. Does it work in is? Does it work in HTTP sisters? Are our other servers theirs? Does it work with HP client, like you, don't make requests with HP client? All these things are on the table. There are things that we're like we're.

B

Looking at we're starting to explore, I think long term. It is our goal to support these things in all of our our places. What we're working on right now is we're playing around with prototyping HP 3 in kestrel, we're not trying to try and implement quick because, as Justin sort of mentioned, you're trying to implement reliability on top of UDP. That basically means reimplemented tcp and that's a hard problem. There's a reason why most operating systems do it for you yeah yeah, so some.

C

Of them Mike just like the team, it's writing em as quick right now. It's like a team of seven or eight people and they've been working on it for about two two and a half years, yeah.

A

C

That's been solely their job and like like we have engineering resources, but implementing that is quite an ask, especially for the kind of turnaround we would like, like yeah.

A

B

C

Implement quick within a year: it's not it spares a.

B

Lot of work, yeah, very security.

C

Risky code right.

B

Cuz you're also reimplemented, like you, a lot of the same TLS stuff that how HTTP works but you're having to re-implement in a different way on top of quick, so there's security implications on stuff.

B

So so, when Justin showed that code, you saw this use MS, quick line, that's because what we're doing right now is prototyping with a library developed by Microsoft called MS, quick that library right now is not publicly released, but they're working on on figuring that story out and that what that library does is it lets us and it handles the quick stuff for us.

B

It lets us, do you know it handles those streams and stuff, and then all we have to deal it on top of that is the HTTP 3 stuff and that's sort of our approach right now we're looking at that, because that gives us that leg up it means we don't have to start from Ground Zero. We can, you know, build up from something that already kind of works. I.

C

Remember, Barry dorrance walked up from my desk and was like Justin. You were not implementing quick.

B

Specifically, I, don't trust you.

C

Like like the benefits that we could potentially get of being able to write, donek code and then like having a little bit more performance, context is kind of out it's out late, but security implications and also the amount of time it takes. Yeah.

B

Yeah, so one of the things you're gonna start- you are gonna start to see in the next couple weeks is we're gonna start checking in some of this code, some of this code that implements hb3. So if you're, following the asp, nut core repo you're gonna start seeing hp3 stuff come up.

B

The context I want to set on this is: it's still super experimental think blazar way back in the early days, and not even that, like we don't even know exactly where we're going with this because, like I said, the spec isn't even done it's they're still discussing this yeah, but but you'll start seeing that stuff. We're because we're using this MS, quick library, those hb3 bits.

B

We start checking in early on I'm, not gonna, work right away, because you're gonna need that library and we haven't figured out how we're gonna deal with that right now, but long-term. Our plan is to devote to produce to some HTTP 3 scenarios and get them out in a version of.net core, that's full open source. You can contribute you can build everything you know follows: are our usual patterns, but right now we're sort of in that early, prototyping and basically, what it comes down to is Justin's tired of private Fork of a cotton. He.

A

Wants to get out of that.

B

Party and and we're tired of it too, we've only been it's only been going for a couple weeks but like we want to make sure that that our custom that our community can see what we're doing, because that's really important to our open source, even if what we're doing is still building on these components that are a little harder to access. I, think it's important that what we're doing the public yep exactly.

C

So John I saw a bunch of good questions. You wanna go yeah.

A

Yeah and I've heard my offering a bit here so feel free to jump in. If you.

C

A

C

Screen and all I do, if you can scroll I was thinking you can scroll, that's fine, oh yeah,.

B

C

See? Half the good comments. That's a promise. Yeah.

B

Our chat window.

C

Doesn't scroll as.

B

A

So there's one from Michael Powell with architectural from connection-oriented. Does that impact request response sender, receiver, listener role, kind of questions the.

B

Roles are very similar like so it's still quickie as a connection-oriented protocol. It's the same just because it's implement at the top of UDP. It's still connection oriented, because when you look at it IP, which is what all of this is based on, is also not connection based. It's just sending packets and so quick is you you can think of quick as implementing TCP on top of UDP, but doing it in this like multi stream way.

B

So you still have a listener and the listener accepts new connections, and you have connectors that you know you have clients that connect to to the listeners.

B

There are some quirky things about quick, quirky, quick that, like quick, supports a feature called connection migration, which is so in a TCP connection. You, the connection, is defined by the two IP addresses and ports on either side right and the local address and the remote address that and you can't change that. So if you go in a tunnel and you lose your and you lose your internet connection, so you connect to the local Wi-Fi that TCP connection is gone because it's a whole.

B

You have a whole new IP address, you're a whole new thing in quick. The connection is not coupled to a specific address, so you can actually migrate that connection so that when you go into that tunnel and you go on and you you know, you're on your bus with wife or something and you connect to the the Wi-Fi on the bus, you, your TCP connection, your quick connections are not dropped, they just continue, as is yeah.

C

And I remember seeing another question it was about like why are the big benefits of a ship III, so that was one of the right.

A

C

Connection migration on the other ones, connections, establishment, I'm, like the time it takes to establish connection, is like especially a secure connection is significantly less um also this head of line blocking problem with TCP and being solved by KTP. It's like probably the biggest one I would say: that's someone I'd, like you know, if you're likely, reading like oh, why are we getting right? Baby, that's probably the biggest reason, and then yeah after that, like it probably is gonna be more performant. That's the probably the biggest question that, like is on people's mind,.

B

What are the interesting things you get out of moving to UDP as a base, and if this can be good- and this can be bad- it's just different it's. This is the way to look at it is today. Tcp is implemented entirely in the operating system. Kernel right, it does all the work and sends you the data.

B

Udp is as well, but UDP is pretty simple, like that's very low logic, but TCP has buffering and flow control and all this stuff, then you can't touch that code in the kernel, because it's it's it's highly got to be highly compatible and all these things quick, because it's implemented on top of UDP quickest implemented in so when you, when Justin, ran that code. All of the quick code was running in his process in user mode, not in the kernel.

B

Now he was running for those DLL that we were using, so you didn't write it, but it can iterate more quickly. It's not tied to your OS version. It's it's more flexible! That way, it's like app locally and this you know in that way: yeah yeah, all right, I I've, seen a couple questions and on a similar theme, which is around platform support. So first off the question that guys since this is built on is quick. Does that mean Windows only well, there's a good reason. It's called MS, quick and not something like win quick.

B

Our understanding is. The plan is to take that cross-platform. So the thing I want to put out clearly is that we are committed to building a cross-platform quick, implement a like quick support and htv3 we're still looking exactly what that means and how that works. We're working with the MS quick team to help them sort of understand what we need and then what can do, but but they plan to be cross-platform as well and.

C

The thing that I, like I, just want to remind our customers, is like fundamentally like if we can't deliver in the fact that, if Emma's, quick Desai's only to run on Windows, for example like we will hit it and find a new quickly.

A

Yeah yeah, like.

C

We're gonna we're trying to promise everybody that we're gonna deliver a cross-platform like great experience with HP 3 and until we get to that point, we're not gonna ship. It really and at that point like we're, gonna make sure it's cross platform. So if Emma's quick only runs on Windows, which from I could tell it's, not it's gonna run on budget yeah.

A

C

Of later platforms, we should be in a good spot, but again that's very important to us all. Right, I think this god this is oops you mind. Switching back to my computer.

B

Yes, I. Remember we changed tonight is: let's try this one yeah.

C

And this is I kind of kind of went over this, so this is generally just like. We have bedrock in kestrel, so we're gonna develop a package just call like Microsoft based for no connections, dot, quick dot, whatever doesn't really matter. Unless we are better a component that allows us to get something called a connection context I, would say, connection context represents to pronounce an imprint and output for your connection. So.

B

And, like the other transports, we have right now that are like this instead of transports that quick, we've also got transports sockets transports libuv. These are the things that implement the raw bytes going back and forth on the wire yeah. It's a little.

C

Bit interesting with quick because most of the work you want to do or actually on streams, the connection is just kind of like a right.

B

You can't send data on the stream right you have to create are on the on the connection. You can't send data on the connection. You have to create a stream to do that. So.

C

We may need some work with bedrock in order to figure out the right level abstraction because we really haven't had this like multiplex, like we want streams instead of connections kind of idea. So we just need to figure out those problems and then I just mentioned quickly that we have Microsoft based Finnick orchestral, that's gonna be implementing htv3, and then you know we'll see what we do with other things like signal are or zero PC. We don't really know yet, but I'm gonna show you quick. A demo about single are using quick.

C

Now, let's see if this actually works, because that is.

B

Something I said and and so date you set this up. So it's like david. I remember, I think I don't know if he showed it on a stand up, but he's definitely talked about it before a while ago. He he liked to show this sample of sigma, are running directly over a raw TCP connection and that's because sick. The way signal are is layered the actual messages that go back and forth your hub in vacations and things like that.

B

We send them down a WebSocket or we send them through long polling or through service and events. But that's layered, like that's just one of the transports. If you want, you could run it directly on top of TCP and so Justin and and David got together and kind of said. Well, okay, we just built this quick thing and here's the thing signal, our imagine you can imagine each invocation could get its own stream. You could have long-running requests.

B

We have streaming support in signal R, which lets you send multiple messages, but it has this head-of-line blocking problem where, if you have a really busy stream, it can block other things and if with quick, you could get some of that raw power right out of so.

C

As I was doing that I actually so producer, this is just an echo server, but it's using single R is sending whatever text you write on the screen and then it's responding back so I can show some it's.

B

Another one of those it still works.

C

B

There's not Minoo, there's not a lot. That's different out of the which-which okay, someone just asked to Wireshark it Justin.

C

Wireshark it all right, so, let's unfortunately,.

B

One thing we're gonna see: he's gonna set this up because quic is always secure. You're gonna see a bunch of Wireshark and it's gonna be a bunch of encrypted content, but you will see that there are quick messages coming in, but you won't see the content of them and because we're sharks, decryption logic, doesn't work with quick either.

B

C

Right so we have UDP looks like it's. You.

B

Know you're on the ear on there am I on the right, so yeah go capture, stop go to capture, I'm, sure the options, let's make sure you're on the right one. You have the loopback yep yeah.

C

It's coming up, it's UDP who's! Coming! That's quick earlier I downloaded a new version of Wireshark, so maybe that affected it, but these are yeah you're. Shutting these are showing EDP packets. They should be quick, though, do.

B

I start the capture again and run the demo real, quick, yeah sure, saving and stop it. Let's restart the processes just to make sure.

C

B

Start up the sample, the client, you started, sir. This.

C

Starts with the client.

B

C

Yeah yep, so there's ood echo.

B

Yeah, let's go here.

C

Okay, there he goes quit there. Yo I saw.

B

C

There we go yeah I, don't know why I would.

B

C

Yeah there we have satisfied.

B

C

B

C

Demands for proof, I know, yeah I was gonna, say no one's gonna. If I didn't show quick at all yeah.

A

C

Wasn't I there may be a way to get raw data out of the quick connection, yeah.

B

You can get some of the data out, but a lot of the pay that you notice, if you actually like click on that and then double and then open that up, you can see it says like protected payload and it's like it's a bunch of garbage cuz, it's all encrypted, but yeah. You can kind of see that but yeah so like this is sort of an experiment. We're moving down.

B

There's just I just posted this morning, I posted an announcement to the ASP, none announcements, repo sort of helping set the context on this as well, and this is kind of our you know. It's our explorations we're trying to look at we're trying to see what comes out of this early on, because we don't know what its gonna be and how long it's going to take.

B

But we want to start talking about it in public because that's what we do, we we try to do our stuff, open-source, yeah and and in public as much as we can yeah.

A

B

I think I haven't yeah I, don't know John. Do you have any other questions coming through you've come.

B

Let me take another look here, feel free to send your questions out. If you want, we still can yes. So.

C

One thing to know is that we're putting all this code open-source community members will not be able to run htv3 because.

B

Of the Emma's quick library we using right now we're so trying to figure out.

C

A bunch of nuanced things we.

B

Expect that to be pretty temporary, where this is just like I want to make sure that people when they go and they clone the repo and they try to use this hp3 stuff. They understand why we're not it doesn't quite work yet, but but yeah and actually just yeah.

A

Go ahead, I'm! Guessing you probably can't answer this, but it's an MS, quick library, I'm curious where else at Microsoft. If there's any there announced quick usage, so I mean.

B

The MS, quick library is developed out of the sort of same team. My understanding is the same team that builds things like h-2b, sis, the HTP implementation inside windows, so HTTP 3, support in Windows and I is an HP says. This is also I talked about now, all our servers, this was in Kestrel. My understanding is that's something that is on their radar. That.

C

Server open right now, I, don't know where it is, but it's with is and they're running HP 3 on it them.

A

C

Or with the quick as well and that's using MS, quick underneath it so yeah.

B

So like first so when we're talking about this we're talking about implementing this in Kestrel in order to get support for this in is or an HPC server, you're gonna need to run whatever Windows version has that functionality, when that rolls out an MS quick, is the MS quick? Is the library that's being developed for that purpose? It's for common code to be used for quick across, not just the company, but also you know, they're looking at moving it out into the community as well and.

C

I know that there are some teams at Microsoft that have started using or had used em as quick in the past. I can't remember them honestly, but there's a good number of teams. Yeah.

A

hmm This is I mean you know, I think you went through in such an organized way and presented everything. I think you feels like you've answered the questions here and.

C

Think of something yes, yeah,.

A

One more slide.

C

It's really like what are the so. This is all nice and shiny, like where the big problems with quick and HTTP 3- and one of this is that people don't believe. Udp is a great mom. I would say. I would say this like a lot of companies on routers along network equipment, kind of walks UDP by default. um We kind of believe that it's going to change over the future that people will realize. Oh quick as a protocol. We need to start allowing UUP again, but a lot of people did it for security reasons.

C

They didn't want to accept any UDP traffic at all, except for yeah.

B

Like I mean one of the things one of the things I was thinking of just now is like when you go into your as your VM or something and you set up through your ad, your firewall rules and you turn to say, I want to allow HTTP. It allows TCP port 80 through well, that's not gonna work with quick, because you got allowed UDP port 80, so I mean I. Think a lot of this is also you know. There's gonna be such it's early on in this process.

B

There's gonna be situations where quic is not gonna work and fortunately it's been designed with these ways so that you know the server and the client can both kind of try to use quick when they can hb3 when they can and fall back to hb2 and fall back to HP one like all these things have this model to. Let you do that. I think that, but but but it's definitely still a question mark of like is UDP the right choice for this. This is why it's still a draft spec. It's why yeah I mean.

C

I the question most likely people will be it because there's been so much so many problems with implementing or adding new protocols. It's only tcp and UDP like there was ideas of like adding like as something something else that was yes.

A

C

It didn't solve all the problems really to no success. They people tried and there's been. No success is just TCP UDP honestly. Another thing they're concerned about is that you need pieces like not very common scenario, so the performance within kernels is not great yeah.

A

And Windows is not that.

C

Doesn't have great UDP performance, but I really I just heard that well.

B

And it's tricky to because of course, UDP will always seem somewhat faster than TCP, because it's dealing, you know it's not doing all that buffering flow control that stuff, but but if it came down to it like it's also not as common a scenario because everything runs on TCP, you know almost everything runs on TCP I'm sure there are plenty of people watching right now who have used UDP for stuff, but like the vast majority of things, use TCP, so it just doesn't get the same attention, but it will allow us to send it probably hopefully.

C

Yeah and then finally like people and saying it's not a big enough improvement over HTTP to performance-wise.

A

C

What was the RPS of it? We haven't gotten to the point where we were measuring per file, our side yeah.

A

C

It's gonna be critical to make sure that HTTP 3 is competitive with HTTP 2, if not better yeah, and if it's not that case, then then, what's the point means supporting it. There's other benefits to get out. It should be 3, but um it we really would like to make sure the performance are at the RPS is comparable and.

B

That kind of gets into this experimental thing, because we have to figure out how much it's gonna cost us to build this, as in like how much time and energy and like effort is gonna cost to build it, and we gotta weigh that with the benefit, which is what's the value.

B

If this is only useful for the googles of the world, the Microsoft that you know the asher's of the world, like maybe it changes the equation a little bit, because I saw one question go by which was like what is the actual benefit to developers of this. The answer is like performance yeah mostly, um and the reality is, if you're happy with your performance in hiv-2, that's not as exciting I mean it's. It's definitely nice to be faster.

B

It's always nice to be faster, but like it has that sort of performance is the core thing, and so, if we can deliver it with the right transfer performance, that's nice. This.

C

Is a feature that we believe is not something that people are gonna notice immediately. Nor is it something that people necessarily want, um but down the line we believe is gonna, be really beneficial to anyone. So similar I.

B

Mean and it's a lot like HP like HP, adds some little bits of functionality. You've got pushed promises which most servers don't even really implement now holy and you've got some things, but, like the major benefit, is it's a performance thing and you can just you just throw a switch to say well turn HP 2 on as well, and hopefully some people get faster connection.

A

um You know I feel like the performance side to it. There's multiple aspects to performance a lot of time. So one is what you know. Ok, my I hit a web site. I get something faster, that's great, but also scale right, and so, if it's more efficient on the server I can serve more requests with less resources, pay less money. All that kind of stuff yeah.

C

No that's true, yeah I, remember seeing a graph come from something I must quit folks and they said that they they were seeing comparable forms between HD 180. Oh sorry, HB 2, +, HB 3 with regards to connection connection opening when it was a very low number of work on the machine, but once it got to a much higher scale, it became very clear that, like quick was allowing connections to get through way quicker, right, yeah, so yeah.

C

A

Let me see there were some other questions, there's one jumping back and forth because rich and everything yeah, sir laser mr. Magoo says, will allow for migration will allow for simultaneous trains over different connections. So.

B

It won't allow us I mean you can have multiple connections, but over the course of the mic. So quick is designed for this connection, migration and and while you're migrating, it is set up to allow your concurrent requests to continue going as much as possible while you migrate from one address to the other. Of course, messages you sent with right as you went into the tunnel, won't arrive. But when you do reestablish the the new connection, it will retransmit them and it will just act like a dropped packet, basically yeah. So.

C

With HB 2 and H 2 P 3, like it's very normal for like browsers or you know, whatever client, you have to only create one connection from client to the server with HT v 1. It was you.

B

Had to create like five eight one, five at.

C

Once, in order to make sure they didn't.

B

Have this multiplexing.

C

B

That was their way of doing the streaming. Basically was they had T's five TCP connections instead of and that's how you did, you could send five parallel requests.

C

And fundamentally, connections are more expensive in streams right, that's the big thing. Is it having five active TCP connections or is way more expensive to both the client and the server than just having one and and because you can multiplex it? You should not have to worry about having this blocking issue. So typically, it's just one from one one connection from the client to the server and then during this reach an admission phase. It may be I'm not quite sure yeah.

B

Is that little as it's migrating over but yeah exactly it's still one? It's still considered one quick connection, it's just the what they basically did is they kind of virtualize it over like? This is what you do and you want to make this thing work. They made the connection, a concept, it's not real connection is just an idea, man and and and then the actual physical UDP UDP doesn't even have connections the actual physical packets. That's just one one instance of the physical transport and the rest is kind of more virtual.

B

Over top of it, no question.

A

From John so will TTP 1 1, & 2 Oh connections be slower if we're defaulting or trying HTTP 3. First.

B

No, so the migration isn't based on the migration is based on, like you do hp1 or HP 2, so, like I mean we can talk a little bit about like what's the pattern. So, let's imagine you're a browser. Just pretend that you're a browser for a second, probably chromium, because everything's chromium, the first thing you'll do is you'll open, a TCP connection to port 80 and you'll, say you'll. Do because the user put HTTP in their address bar cuz, everybody should put HTTP in their address bar it'll.

B

Do the TLS handshake so T and part of TLS? Is this new extension called a LPN application protocol negotiation, application, layer, protocol negotiation and basically it's a way of during TLC TLC, TLS negotiation, you say: I want to talk, HTTP, 2 or I want to talk HTTP 1, and that is the part where you decide whether you're doing HP, 1 or HP great ok.

B

So that's what happens today when you connect to websites that support HP, 2 cool, obviously that only doesn't work for UDP, because it's a whole different kind of thing, but you now make your first request. You say: ok get me the main page and in the response that comes back from the server there's. Basically a header and the header says by the way: here's your data, but by the way, I support, HP, 3 and here's where I support it, and you can trust this data that I'm telling you for X X seconds.

B

Like it's a it's a caching things you say hey by the way: I support, HTTP 3, that's on port 80 and cache that for a year or something and then the browser will cache it and the next time you connect to that URL, it will say: well, I, don't even need to do the TCP connection. I'll go straight to the UDP connection, so they have that I think this is a good observation that like well, if I'm gonna try quick first, that's gonna be annoying. Well! Ok! It that's.

B

Why you do this kind of light up? You upgrade upgrade your way forward into the the new ways and.

C

I also believe that sometimes chrome makes multiple requests in.

A

Parallel yeah so.

C

It has knowledge that there is a quick, quick end point it still. What tried to do both a tcp and a UDP you want and see which one comes back. First, cuz I've.

A

C

That the quick one is slower for whatever reason who knows maybe well.

B

And that's the other thing is because it's tcp and UDP are very, are often very like separated systems in in the operating system and in sockets you can just sort of like try both yep.

C

Yeah and at that point, I would know which one's faster and then it making and.

B

I think you know, and certainly browsers have to be careful about doing that, making sure they're only doing get requests cuz. Obviously, if they send your bye-bye widgets and requests twice down the thing, that's gonna get complicated, but they do that kind of thing when they know that it's a safe request to send they, they just say well what, if I just send both and see which comes back first yep.

A

People are just speculating about different things that you could do like. Could you load part of the page and then load more on scroll or yeah.

C

I mean you can just do that with different. Like I mean it's kind of just a web tactic in general.

A

Yeah, you know they.

C

Had the whole like, like the classic demo day should be to is, like you have a big photo and you split the photo up in like 16 pieces like four by four and then you show like voting with HTTP one. It's super slow and then they.

A

Should be it's really quick like if everything goes.

C

At the same time, yeah um I think the same thing that applies to different parts of your web page. They.

A

Don't want to load.

C

Like haunt at the bottom of the page, you can just wait to do it until I think it's more of a client-side like decision to do anyways so yeah.

B

And I think I mean this is the thing about when we talk about performance on the client-side, HP and HP 3, the biggest thing I think a bigger thing it does over like.

B

If you think about performance is just how fast can the bytes get down to my to my client, then it could be a toss up depending on how well as design but with HP one, if you request that web page and it has to download 20 different images, 50 JavaScript components and a bunch of CSS, and you haven't minified, you haven't done all that stuff because you just yet you don't like this is what we did in hp1, we minified and we merged, and we brought it down to one in HB, 2 and an even more so in HB 3.

B

Because of the parallelism that's there, even though it may not be faster to get each one of those requests completed, they're all happening at once, so you don't have to like I'm, not gonna go as far as to say you don't have to bundle and minify with HB 3 or HB 2, but it is a different you're, making a different consideration at that point. You know you may not need to bundle or minify, because you're just gonna download them in parallel. It doesn't matter yeah.

C

Xuxa Nader asks if how did the same way to be free that will be decided in the future. There.

B

Will be a way to determine, we always have options for that stuff. Yeah.

C

B

Don't even know how we're gonna enable it yet, but yeah.

C

There were some interesting decisions we made with how we enable or lit up it should be to support, wants to figure out yeah.

B

Cuz, like in two two, we shipped HB to support, but it's off by default in three: oh, it's on by default, but you can turn it off again if you don't like it yeah. If you want to be slow and not quickly, yeah yeah.

C

So most likely we would make htv-3 on by default early.

A

C

Version and probably would be in a major version that major version I am uncertain: death yeah.

A

C

The amount of work there's a lot of work to do still. um We don't know what necessarily is the bar for us to running the ship. That says like fully supported, but I'll tell you. This, like I, probably would be working on this pretty much dedicated for at least a while yeah.

A

It's a focus for me, I know, there's people that.

C

On that side, they are very interested in well I'm, trying to figure out a complete story across all I've done that yeah.

B

The site we haven't talked about yet and is, is well what about HP client cuz. This is great if I can get it from the server, but like and chromium supports it. So that's cool I got my edge and I got my Google Chrome and Firefox is working on it. All the browser's are doing it well, what do I do if I want to do service to service communication using HP 3? That.

A

B

And the most I'll say there is the that's on the table. We're looking at that, like I, think we definitely have support for it in HP client, as we go forward again same same sort of caveat I'm putting here, which is that this is all very experimental we're playing stuff. We want to show you the stuff that we're doing early on, so you can see how it gets made and how we work on this stuff. So.

A

And you've presented what you know your early thoughts on this and there's the announcement repo with a link to discussion right. So that's where people can talk about their use cases or whatever yeah.

B

Yeah and so one of the things, if you go to that announcement, there's another link to an issue on asp net core. This is what is an epic. This is kind of you know. People who do agile, planning know kind of what an epic is.

B

It's basically, our big roll up item of like the things we want to do right now, I put like I put like three things here this morning like we're still this is you know it's still a lot of speculation on like what exactly do we need to build, but we'll start kind of adding that in and and like keep fleshing this out and expanding it and making it get a clearer picture of how it goes.

B

But but you know we're gonna start checking code in just because there's code there doesn't mean we know exactly when we're gonna ship it yeah.

A

B

Gonna keep coming back talking about it more often when we have more ideas, we're gonna, pull blog about it. We're gonna, put announcements up, we're gonna talk on the stand-up like we're gonna talk more as we get more idea of where these are gonna: land, yeah, okay,.

A

Well, it's been just over an hour, so it's probably good spot to wrap up here. So thank you very much, of course, we'll have to get you back in as things get more shaped up and and then I'll get your get those links from you so that we can add those into the link roll up that I sent out earlier yeah.

C

Great yeah yeah again thank you for having us on John yeah.

A

Yep. Thank you very much.

C

Cool we do, we do a dramatic zoom out.

A

B

I do have the button that the big like and everything but.