IPFS IPFS þing 2022, 10 Aug 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Planning workshop for IPVM

Description

----------------------------------------
Missed last year's IPFS þhing? Register to attend IPFS þhing 2023 in Brussels, Belgium here: https://2023.ipfs-thing.io/

A

Execution boundary.

A

A

Okay, okay um blossom or not.

A

I feel the base component.

A

B

B

A

Right because we might have more context at the end of the day, or it might be more of us breaking out groups or something uh beginning right at the end of the end.

A

A

Three four.

C

D

And so on, so how do you package inaccurately.

E

A

Okay, grandparents hypervisor.

A

A

Okay, great, so anything else awesome, so we're going to start with, I believe, color convention interfaces and move on to execution, bounding, ipli hypervisor linking marketing as well.

C

Do you want to have any norms for either taking notes time boxing or just let it fly whatever.

D

Again, community can decide, I think we should um and we need about.

A

30 minutes from the time yeah, so we'll just call this end: minus 30.! We just stop when we do that.

E

A

Then sorry, how much are we talking about.

C

So let's say uh with that in mind: let's say: do we want to give any order time I mean yeah.

D

We talked about 25, we targeted 20 minutes per. That would be a good size. Something will take longer yeah.

B

Okay, I'll do.

C

Please do take notes everyone, let's make a joint thing.

C

Yeah, can you put it.

F

B

So, do we want to draft of what the fbn is doing here for the most conventions.

B

One con convention is between factors uh and they're, pretty simple. So basically, first you write a block of the system or blog system.

B

Don't actually it anywhere? You just write it. Then you call send on the target actor id handle. Then it spawns the other actor.

B

uh The entry point is always invoked.

B

B

If there is a return value.

D

B

The function signature is always instead of argument and the return value. That is, the error, number function um if there is actually an actual return value. The return value is the first element uh of parameters, so it isn't possibly pointer to where the terminal should be written in memory. uh Then you pass it on the parameters.

B

uh Then we would return.

B

And if it's non-zero, you say oops, it says: if it's zero, then you say like did you look at the whatever things you told my memory? Let's say that was hard to really understand. Are those.

C

Just conventions in your components itself or are they broken out of any kind of proto or anything.

B

No, that's a convention that we just have. um We documented in the sdk.

C

Just paste in there.

B

um Yes, uh multi-return values and also are kind of great uh that may probably grant you there was a shot. It wasn't fun, russ, the russian tyler and albion and said, like we weren't.

E

B

Hell all the time uh where, like someone was being helpful for turnover, um so the initial version of this had like no parameter.

E

B

Alignment is annoying.

B

We had going with uh packed representation in rice, basically where it's like. You have this new truck and you take all your fields. We pack them together, which is possible in the padding and manual alignment to make sure that everything works out really well by just like you know, working in the fields, um uh but that was some of the best objectives.

B

F

Or is it just like you.

B

Know it's like we. We chose numbers as errors uh because it's a blockchain we didn't want to have strings.

E

B

Wanted this to be very like deterministic, but we want more information than this one we would say like hey. The parameters would have grow.

B

um uh We give a pretty good set of numbers for this, um but they're pretty specific system as well. uh Basically,.

A

D

A

B

At the clinical dimension at the end.

C

When you think about, if you were going to apply this to non-blockchain scenarios, is there anything you would change or generalize so.

B

We don't know a column between um between actually models, um uh which is kind of like so, except for so like the sending is more like at the level of actors or several objects, but you want to go into the libraries um our like the basically kernel, wouldn't work so well, that's specifically optimized for columns, but we actually here's like we call it called where in our loss memory we have things. The problem is.

B

Until it's done using it, uh the nice thing about graphs is, we can talk like times here and guarantee that they can't accidentally hold onto a topic that references data. Beyond that point,.

B

I'm interested in.

B

Sequences, okay, I use the height uh the interface, that's regular, positive system. um I use um like just like c-structs busy tech that has all these different options uh and just create a network. Speaker like this is how fast they are when it works.

C

If you were going to.

E

Based on your learnings and implementation,.

C

So far, if you were to redesign it from scratch today, anything dramatically different.

C

B

What I know this is what I would go with because, like we've already used myself, the first version of this kind of turned on the second version should not be used afterwards, which is a problem with blockchains, um because it's just random data, that's unspecified yeah, uh so like basically, we have several iterations in them. um Well, one thing we didn't realize was that at least for us calling them out of blossom was not as cheap as we thought it would be.

B

I don't know why it should have been cheaper, um but something is was seemed to be expensive. um We don't know it was a fashion, construction, cash or if something was happening, but we found it just like it wasn't.

E

B

About so, we actually had the application into the carnival and back into the blossom uh protein infection and metadata information we used to have like one called perfect testimonial. We now actually just like, have a single big structure, just copying them. uh Another thing we actually would like is some way to vote. The other thing we need to do here is we.

E

B

Have any any uh management calling, so we don't call it into the kernel and have it kind of called back to us uh yeah. I I I like that. On the other hand, it's in the money because, for example,.

B

I have to like, like, for example, when we open them, one call to open it, get a handle and the size, and then we have to cover internally and we call it and say please or you just frontal type thing uh would read and not just give me the data. So if we.

G

Had to generalize this.

B

Out of just linux,.

G

Because it's going to run in all places, oh, but.

B

That's right, I I copied, like all.

B

B

But that would be probably a massive boost, but then you have to basically define what is your apparel latency is parallelism.

B

I I guess once we start getting in on the charts and some of the threads that's up and yes.

B

Yeah there's the biggest frustration is, is actually both with a couple more we're trying to get it, for example, because we have to know the size of where we'll begin. First.

B

Having like how we can copy data, it's kind of the ones that we have yeah, so we do have to copy data and like reduce this as much as possible.

F

um So the host you can read like into um the inside of web assembly, so you don't have to copy out the load assembly as long as you just repeat it.

B

As long as it only comes, it's only only reading it for this current function. Yes, the other thing here is like, uh like I did probably in I really really wish. There was something like that. uh Some section of memory wasn't and say like hey: this is the system reserved number? You cannot write to it, but you can read from it.

H

B

Most modern, actually all modern, obviously do this, uh and this will let us expose things to the possible module in like a very like business, where you can't touch anything, so this would be a great way to expose all the blocks or, like all of the um uh you just have to use modules.

B

No mapping is it's like support, but does not exist like technically there's response to multiple memories. So it's not the eyes on that side and there's a lot of questions there. How do you have to work with that? Because that's like computer archive, like snark, because you don't have multiple memories.

F

Importing memory is not the same thing when you're linking right so.

B

You can import a memory or you can export right. I still don't know why people do both of those right now, um but.

E

B

Problem in theory,.

B

Will be specific to memory which is really cool, but on the other hand it means that, like I I I can't pick with memory when you just have to think.

B

Multiple memories: they just accelerate into a direct space yeah. These are entities.

B

Okay, so we've got six minutes left with this topic. um Do you want to share yourself? I want to hear about how like vision, works.

A

If we started like three months ago, so yeah yeah.

F

Yeah we haven't done much yet so we've done the button button thanks and that's it so yeah. We need to look.

B

I

Previously, it wasn't quite very well, probably maybe I should describe something about like interface types proposal state now. I think it would help for this for seven issue. So in three types now, so they are revised only in washington and then the folks from austin time they follow the most easiest way. They just say: oh, we have like three different types and we have so-called canonical civilization.

I

It means that they don't now have uh instructions for interpreting types or leaps in the whole brain and so on. They just say like this function takes string, it takes away and the string is utilized. String is encoded somewhere in three types and like then they say like we have canon dot, lift function and it takes arguments functions that.

I

And according to that, they have like for each type, they have a strict rule, how it should be lifted over. So that's how it works. Now in the obviously specification in a proposal in three steps, and uh then what we want to do is uh so-called adapted function instead of like refined communication like.

I

They would like to have instructions like special set of instructions that intended only for mixing and all real times and.

F

I

I

Because uh for in each certain instruction you define your own type and you can define your own constructors instructors for each type. I mean like to devolver and.

E

I

It's the main point here you compile from different languages, like imagine that you have string in c. That should be multinational and raspberry should be utv.

I

In current proposal button like in adopted function, uh there will be special instructions, they know because gonna become power of assembly in the compound, and you know everything. Yes, we know operation system, language, uh common convention and stuff like that and during calculation you could choose instructions that put like lower elliptical types and even like we described yesterday.

I

uh I think that uh so it's not not an articulation, but I think if you try to work with the web assembly, so.

E

You could try to have.

I

This, like simple function, just or something like that, that checks that types in different modules, uh they're compatible and.

F

E

I

You could just copy them from one memory to another and just like, which is.

I

Structures and so on.

B

I

This pro, like a company model proposal, has very difficult faith. uh It was firstly uh like proposed in 2015.

F

There's a lot of companies that are specifically looking for it for optimizations.

I

E

F

At least the component.

I

Model sorry yeah, we could present like we were expected on plus two awesome time uh video closed and, as I understand guys like you know, in the previous month, they decided that they uh like they have been one for this week and the reason we can just try to count.

I

And now they understand they try to push.

C

Just before we move on, we want any action items out of this, or should we wait till the end of the project to play with the manufacturer.

I

um I think I would recommend quotes from uh try to like come to overassembly folks and try to uh say that there is a demand from our side point of view. Steps, because uh I would say that if there would be a lot of exceptions,.

I

C

You said, there's already a tracking issue right, yes, yeah. Thank you.

I

I mean that that would be the avenue.

D

Okay, moving on how's.

D

To see how they do all the calling implications and hanging of promises and references to objects and callbacks and whatnot is very javascripts oriented. But.

H

D

A

A couple paragraphs.

B

B

Parts to this, uh uh here's describing the interfaces from muscle components. This is getting into the uh matrix types. Language um there's also describing interfaces, uh but the important part here is again like calculate these side of things. um uh So, like you need to describe my interface, but also specified like an ideally data structure that I expect they could link to other things as well. um So we have a couple of options right now.

B

I know so I know we have um uh what's called.

B

I believe these chemists um there's also.

E

B

Working performance project called uh always.

H

B

B

Yeah, those are the two of us. We have again, you guys have uh other base languages. You.

E

B

Using you were using.

I

We need some way to describe them in the different languages because, like in glow, and uh there are more procedures- and you can't do it like it's easy so and uh now uh so like one also uh like there is another project called infinity and phoenix. There is a special language for candidate in this game. It was also inspired by white. These types, uh the author, one of the foster candidates, is the one who was initially proposing three types into assembly and they have quite even they have proximization complicated, I'm not sure yeah.

I

So and I I would say that they're very important uh when people start with us.

B

It's reasonably extensible um because it's something they need to do on and also additional kind of data as they they need it because they feel like you might want to start expressing like different capabilities. You need to calculate this or, um like add the additional metadata about. This sync is guarantees all this kind of stuff, um and it's in these languages, where you don't have too much. That's like yeah, then you're, supporting whatever you see, is that functionality. um The other thing that kind of we'll talk about here is that they have not.

B

uh I really want system for negotiating databases, um um but this is. I don't really know how to approach the topic right now, but.

E

B

It comes down to being able to describe a set of interfaces um and then people somehow name those interfaces in some way.

B

You just dropped the library in the name's hash, but they've also been a way to actually figure out which interface you want to use and how it's implemented. So if you just have your interfaces that doesn't meet you everywhere, you need innovation. Then you need somebody.

B

Entry points, but the main one is you just invoke data and data out. um Sorry, I'm hearing a bit, but I guess my point is like like having a little script, really just one thing, but then you also need something to describe how that maps like actually calling a module or you might was a little more flexible than just like the pre-defined physics.

F

Is an interface that negotiates other interfaces um like multi-stream, select, basically or.

B

V2, it's it's somewhat. No, it's the same idea um but like yeah, um when you do that you're gonna write an arbitrary code, uh so you can do reflection based nothing.

B

We need to not only think about the interface description, but also the internet actually want to react, column dimension um and also consider things like dynamic calls and stuff like that. Currently, basically, where it's like, I explored this exact same function. It's not like. I have an object that really.

B

Basically create more objects, but those are still running with the same servers.

B

How do you describe that kind of world when you only have like one module? That has a few point points that you need to be able to say like. Oh I'm giving you an object and it had it it's like sort of like basically here's an id for the object uh and here's the interface for the object, but to call into it is to call back to me. Tell me about or tell me the objective method. um That's also, I think, something's not covered.

B

This makes sense really like you might have some some instance there's a modern instance. We have single differences, but actually multiple objects are instantiations of that and they're all kept internally.

B

So I just want to talk.

F

B

Yeah, these are probably but there's different types of suits, but, like types are limited, I know that male wise is like basically effect system, uh so it has um unions uh effectively some types of news, new types and a bunch of effectiveness. So all the difference you might need- and what was one of them.

I

Because it's like it's very good for a conception of encapsulation, not.

B

Give you some object, did things kind of point to something back in my world and like describe this type in some way, so that, like you know this, isn't powerful stuff like that, but, like then like, when you run into this this type of your system, you should then call back into me. So this is for. Like the facebook, I have a massive map on my side but includes object, references, it's really nice to be able to do that in some way.

B

Basically, remote.

I

uh Like I would say that uh all this problem regarding comic conventions at all, they will be solved if uh the last version of infrastructure will be released, and the only thing that we need to do in ideo is just to make this ideal itself to describe types of uh in unique language of functions that you want to export in inverse and then, like. Probably, this idea should be embedded into.

D

The thing that one was using to define all their interfaces.

B

Awesome execution boundary.

B

So this is generally understandable.

G

You um figure out it's some difficult stuff that like cancels like timeouts and things like that uh or you may want things that involve. You know more comprehensive, like how I manage.

G

Trying to like linearize all the different add them somehow or something else, but you it appears.

E

You did something, and as soon as you have dynamic.

G

Linking you like really really need something because and then you add user dynamic linking and then you really really really need something um and there's sort of like two two phases of this um personal.

A

E

G

Like how do we want to limit it and describe it, and the other is how how we want that to exist? More broadly right, so you know, let's say like the perfect gas model. We can linearize everything and remember it's like I'm.

E

G

With like 15 units of gas per thing, um is this like global? Do we tell people like? Is it a blockchain environment which is the gas stuff has been happening? Everyone is is like supposed to be. Like homogenous, everyone gets 15 units of gas, otherwise the system stops working because, like the determinism.

G

That may not be our environment, and then we have. We have that.

B

Well so, like most boston run times, it's nice for, like the boston time, has fueled me ringing.

B

Version because they don't require the exact metering, they also have memory limits, snapdragons and stuff. uh So you can like you can even use pool. So you can say, like I have like 50 watts of modules that are in this pool. um It has the size poppers, whatever.

A

So there's a bunch of ways of doing.

A

Get really really sophisticated um gas is extremely stupid, but it works uh and end users will understand, there's a number, so I make them a little higher or lower, or do I just switch it off? It's like I just.

E

Want to run this thing forever,.

A

Cool no more casts done, or I gave it so much gas, and then I ran into the apps. So we just suspend the vm and say: do you want to continue?

A

Yes, no right so like it opens up all kinds of interesting things when we get to ipli after the thing runs, you can say this took as much gaps and then, if you want to check the output and you're also going to run it, you can look at it and say well, this thing took ten thousand gas and my limit is one thousand and you can't run that one of the number do you want automatically. Do you wanna bump the number for this modification? Yes right.

E

A

It it simplifies a lot of things and is probably it's like one of these cases where it's like yeah. We can do all these things. We do uh time date. You know fair timeouts. All these things and yeah gas has a cost cost.

A

um But if you're that performance sensitive then turn off the gas uh and there's also ways of uh you know, maybe you can provide like a timer option. I think, as a default.

E

F

We can take stuff from what we learned from that because we have such calls for like doing certain operations. So we can take some learnings from that.

E

Really, I think it.

F

Just boils down to like you.

E

H

F

See how that is relative to like the rest of the gas in the.

E

System and then just put.

F

A price on it, but so you can have reasonable defaults, at least.

A

That's something that happens a lot in blockchain gas rebalancing because it's like, oh, we thought it was going to cost x, but now there's an asics came out and that's actually y or apple releases, a new architecture, and now it is often disaster in these specific ways or slower in these other ways um that could get tricky because something that used to like we used to make the forums. Look like this, but now it looks like that hardware.

A

um Yeah, that's the thing or you know you know you, you add a new syscall to it and realize, like oh crap, this is actually like way.

E

A

Expensive now we've got versions gas schedules, but that is true.

F

E

Mean would you want a version.

F

Like a global gas schedule that you want everyone to use in ipv when.

E

F

Just like have them define their own, like I don't care about gas I'll just run everyone's stuff for free. Why not so.

A

I think you can go turn it off completely just out of the gas zero or just like forget this mechanism, when I'm running and it'll actually slightly pass through those instructions, um but uh and yeah people should be able to redefine it in their local system, but we need to give people good defaults right so yeah, okay, and I also want to know if I give a gas limit and I'm going to run this thing, and I ran it on another machine that this will run inside the text it's like.

A

If I want to run it again or replicate some of these results. I have added data sets and I want to replicate the results on it with you know like there's two new roads. I just want to run a query. uh I should know like the order of magnitude that I'm running in.

B

For when you get into like deterministic journalism, does this help at all.

A

To have uh well, I uh yeah, if you're running it again,.

E

A

This totally helps because that helps you schedule the drops, but that doesn't help you really ahead of time, because now you have to to to do that kind of automation, optimization ahead, uh you have to do an analysis right and it's basically like add some nurses and stuff. So essentially you start to get around like when you do uh uh ahead of time, cost dynamics you're often.

E

Essentially, doing like.

A

There's some political execution right right uh and so having gas mirrored at right times doesn't help you anymore right. So, but then now you have a startup time. That's like we're. Gonna do a bunch of uh analysis before we run.

A

Is they look at the control photograph right and say these are things that depend on each other? These things can all run in uh in actual uh parallel dependencies on them and then we'll wait here when the flow is joining them.

B

I

B

From a privacy perspective due to rail, but like hey, I'm not getting information about the machines I'm running on. I don't care about that.

A

Well, so that's a nice thing about gas is.

F

It really depends on what features you are exposing um like if you have like. I have these hundred injections that you can.

E

G

F

um And you can probably figure out what they're exactly what they're doing, um but it's like. Oh, I can do this simple, matrix transform. Then it's like well, a lot of pretty scenarios can do yeah, so it depends and it's like warning labels for anyone who's bringing their own yeah again.

I

Yes, I would say it also would help in case single sort of execution, because it has.

I

Like stack and keep like all, your points should be.

D

The guy's model affects the con convention, so if there's any different, yeah description affect the comp dimension across tigers. So similarly.

D

You can have local meaning, but you need to be able to describe the location whenever you're, making an opinion, video or giving it some kind of part of yours right.

E

B

If you're impossible to do that, no that doesn't affect the actual connotation.

D

Well, um it might because if it's not pure and it has some effect, then the effect could be impartial.

B

We're not going to get the same actions, I mean they're securities.

D

um You might you may or may not be able to unwind the whole problem. Yeah, you might have done something to the world.

A

Like here in so like with it's a pure state in particular, uh so yeah pure events require that you should be able to go back in times.

A

um If you have like an actual side effect extensive tweets, then you have to capture the like the description of that happening and you need like.

A

So it's like a completely different thing, but if you're going to then replay the apps, you can say hey give me a new like because now they're actually different, right and there's a new. They have a different hash, completely new indication, new number, let's start again versus I want to rerun this and I'm going to pretend.

E

A

I made the call to twitter, and this was a response.

G

So I need to be able to like, if I, if I'm ever going to do something that is unwindable then like I shouldn't execute it. Unless I have been passed on.

A

uh Well, so you.

G

Partially through it.

G

You should be responsible.

G

B

A

Like if you have a let's say that you bought halfway through and you wrote some stuff into a quantum stress database and.

E

A

Fail because cosmic ray all of that stuff should be deterministic. Sorry in.

E

There's a model.

A

But this is deterministic right. um If you have certain execution and.

E

So somebody should be able to.

A

Resume from that point, like let's say you break this on a module or something you should be able to pick up and continue, but essentially like take the time at the previous tick. But that requires consistency, structures right so like we start to pull a lot from the functional world. Always we're talking about this stuff, which then.

E

A

Leads to questions about so failure um is this: I'm replicating everything totally locally, or am I making calls to other machines and coordinating between them, because then we probably want like supervisors, all this other stuff and restarts right, and it starts to look very early, probably probably.

I

Influencing trying to uh like uh be like in the sense of understanding is about allowing minors or like not owners, just to set limits for their resources that potentially could be used by developers and emotions and they're interesting. You have some similar solutions.

I

Or other machines like two.

E

H

So, in our case, like.

B

um Because, basically, you have two, you know multiple dimensions so like there is a like storage dimension, which is growing over time. uh There is the like: the current storage dimensional resources, that's his name, then you have uh like time, which you always need to ask um you're, also funny that this is interesting. When you have parallelism, you actually care about resources and.

E

B

Sharing them, that's when you want to start charging some kind of thing, but you're not really it's more like if you're targeting them for literally.

B

So so, if I'll go beyond paralysis, there's no like did. That would make sense. If you had a bunch of different processes that could sort of share things. uh Unfortunately,.

B

Because, like um we even have parallelism, we would have very fixed cracks. I think using this amount of statements, it's.

E

Really hard to like say.

B

Oh, you were only using this period to find my friends if you squeeze something in over here, uh you could like, if you were to do this, would you always say you kind of have a little piece of resist where you say like okay, you can um basically target for the area of the curve for how much uh memory and time you're using without extending a little bit. You have to need some time to rescheduling those things like kind of drop off.

E

You have to work, I mean the different dimensions are seems like.

G

Peaceful and I feel like the qp has been thinking about. How do you start playing.

G

And, like I don't know, but putting a scale on it is hard. um You know whenever you apologize. These are all different, different quantities that you can golf and cause failures in different ways.

G

It's certainly I mean, then you can protect the user pane of trying to scale each of these service.

G

That are not doing the homogeneous, blockchain thing, but the heterogeneous.

G

A

With gas but maybe yeah, I mean like another.

E

A

Worse is better right. So, like it's a it's extremely stupid mechanism, but it gets you moving and then so it's like the other, you know stupid, but it works.

A

Distributed um the main question is: how much do you care about if you were able to run something with these inputs? Should that be interrupted too right and gas is really nice for that, so.

E

I think we give people.

A

A

If they're really sensitive to it right, uh questions about failure, and especially if it's networks become more question because now it's like well, this thing failed. I'm going to retry a couple of times. I could potentially like there's a great lamp or quote: that's a distributed system is one where a failure on another machine can cause your machine to fail right. So it's uh like you rent all these property things. You actually have problems on a blockchain.

A

What you have is transactional memory, realistic. It's like I'm gonna, I'm gonna fork the universe attempt this and if it works, I'm gonna make a text consensus. But it's also like this failed 50 times in the middle right, and so the blockchain doesn't have to think about.

A

So having retries with reasonable limits and all these things, but we have like as we network yet we have to recognize that things will fail and take down the good level of computation. Sometimes if we really care about up time, then.

E

A

uh You know really high restart rate and create a supervision, and then it's like this thing went down so hey everybody else. You'll also need to start. But that's what happens when you have a like there's no way.

E

A

You start getting into like smarter programming models that are um like where things can get merged after the fact right.

B

Next, possibly.

H

A

Of describing the job that you want to run so here's.

E

A

Lawson blog container, whatever here's its arguments and some other metadata that you might want to tag it with or you know these are the syscalls. I expect to be available and all this other stuff I.

E

Laid that out in.

A

Ipld and then optionally, on the other side of this, I have like this was the result of this session of this job uh and there's a bunch of projects that are doing things like this right um that we should just find a standard way of look at minimum, the learnings from them but like if we could come up with a standard stack that everybody uses. That would be amazing right. So it's like ipvm map file.

A

Fluids, like you know, blog point workers like all of these systems, so I opened up to those projects. If you guys do this.

C

I think that you should go even broader, like I think we could figure out how to do this with build the crd for kubernetes or something like that and make the general implications better.

F

I mean for cloud players just like. Basically, you just have fetch and then that's like a majority of what workers do. Yeah.

H

uh I mean to like call the work, that's like the most used, but it's not the only implication you have that you can like have a crown drop. You can have like that various ways and like from like, if we like, want at some point to mix this with like content. Addressing then like you have.

E

H

Content like how do you invoke the data like on that content for now, like it's still like, rather broad.

A

Also, I think so, just because you mentioned comrades, uh I don't know if you should even get the same layer right so there's a single implication and then like how I want it to get scheduled potentially in the future, but I would love to have a spectrum but also say run this every week or when this event happens. Like I get it, you know, I see ipms update rerun the thing.

F

um Well, it shouldn't be too hard for an actor to like figure out itself when the next point is and then just say, oh at this epoch, I want it to run.

G

And then do that.

C

Are we mixing two different things together, meaning, like I think, there's an orchestrator thing and there's an execution thing and the orchestrated thing says it's time for this and then execute.

C

I fired focusing on the sort of definition of.

B

B

And we need some.

F

I mean if it's a convention or an interface and.

E

F

Were back or back to web assembly, you just have that all as a web assembly module.

E

F

It's not too complex and it's just invoking stuff right.

B

How much of this is accepted by all of the projects, like all those things like pops? Are you guys guessing yeah right so, like that's.

A

Into that oh or set it to.

D

I don't think we should be expecting all this to make his way down, but rather he might yield things that would like it or like use all the underlying components.

D

A couple questions here.

D

There's gonna be two or three layers, so, like one layer is, any locations of a function um simply gets smaller.

D

It's really a message passing from one small object to another column convention: that's wrapped in something that allows you to transport any application across systems and there's something much harder, which is then gonna invoke a whole container or go around a much larger job. That seems to be.

D

You might use the first one in the second um and then there's a third thing, which is you have some long-running state on a thing. You have some object with some number of state and you want to apply some information to that long-running object um which might yield some additional output, some other messages or just a state change on that object, which is close where to the to the particles and the stream of implications that you have there's.

B

A lot of prior arts like the business intelligence space here like that experimentation um specifically, so this is like etl pipelines, do a lot of this project.

F

um Well, we just have like the just straight up invoke and it's a thing: um there's some talk about like for bootstrap actors, that, like is an actor that starts other actors that what it can do, but there's nothing concrete, at least about like, like what parameters and everything that is.

C

You touch on a really interesting point that, as part of respect, it might be interesting to think about recursive calls and basically having at the top level an invocation spec which calls you know, other educational specs yeah. Well, I mean this is a pretty normal thing. You.

F

Have actors are calling other actors, sorry, but like an actor that constructs other actors? Oh, I see a z, yes, so it does a bit of like you could have a chunk actor that you could assemble, especially if it's really large um and how to pull everything and have it. We.

B

Actually, do we will do.

B

Module and just call it that dynamically.

B

But yeah, like I also like static linking um I statically say I like these.

E

B

uh Question, when do we actually visit.

A

But then there's a cost.

A

Just to get like a rough feel for it in in this room, how many people's projects.

F

It's like a 50 50., 15, 15. yeah, whether or not we actually use it.

B

Yeah, I don't know, I'm.

H

Almost like at this point.

B

Like changing a lot of our pictures were working cool.

B

B

E

B

D

If something's very good, we're all getting updated, absolutely.

C

I mean we yeah, we have anything, we have a convention. I would toss it out immediately if there was something else remotely good and affordable.

F

Before talking about.

B

One of the things I just completely.

G

Closed is the next step was: let's go.

B

Look at then open. Let's project, that onto what you can say if you have like does.

E

B

Access require disk type stuff, let's see if they actually mess according to the subsets that you can use. I guess is that okay, spooky spooky deaf.

F

I, like chrome's, like show scary warning blocks before you do something, um but that's not exactly a um automatic code thing yeah.

B

F

E

Like a good model for.

B

This it's obviously like flat packets, like that, like or like your warm water number or something like you, have it after hand, says hey. I need to file that, because this is the topic that technically grants you permission, but it's you know you're doing.

B

Before your communication.

B

E

B

Way to assign things on chain without an idea.

E

And say whoever's.

A

Using this one like a smaller group of people, but they can interoperate so as an example signing ethereum messages right. uh There was a concern a while ago, like well, the ethereum signature format, isn't.

A

E

A

Expect that to be generally available to like everyone in the system right so yeah, you can do like.

F

Well, kind of you: can you can't really like sound.

A

And like the problem with adoption, you're going to have, there is like really big right. The the point in this conversation is possible, even if it's not.

B

Create your you: can it's not really like an extraordinary system and.

E

Then, just by storing a special way.

B

It's now signed by you.

A

Yeah I mean like yeah, you could write the you can into the chain or.

B

The new camera with the session product and know that, like that state, could not be created in any way, which means it's changing right, very comfortable to send this around, because you're anywhere right.

A

Yeah and so like, and especially if you can like like to take like super super like triple underlined, clears like adapt it to your use case and other people adopt it. Then that can be respected. That gets transformation.

F

um So just some sort of like a better thing that I want to kind of expand on is like her.

G

F

That are like dangerous, well dangerous enough to hurt people um like even just having it as an actor for a developer. They might just enable it because they want their stuff to work, and that's just that's all they care about. um So this.

E

F

I, like the rust models, like you.

E

A

F

Which looks really scary, and then it does the thing which is like it does.

A

F

But if you're doing this right, it's fine but the unsafe like. Is it like a thing for you to think about? It's, like oh shoot. What do I need to look at to make sure I'm doing this right?

F

F

G

Requires like access to your bank account in order.

B

I would like to see us as a community spend more time thinking about studying and like really how we deal with the issue like having better.

E

G

B

In practice for numbers- and I'm not really worried for this community but for others like getting the subject, superset stuff right is actually quite difficult.

B

Like one of the big things we should consider here we're singing about hey. We have to provide a capability to navigation function, there's going to have to be a check to some sort of.

A

uh Delegation eventually consistent, uh except for the like expiration right on it, which is right. So it's.

B

Proactive yeah, so you can, you can execute and find.

F

A

Yeah but that's right, it's like uh if you read, if you revoke that's your last possible line of defense, so don't think of the like, and then I.

A

Well then, yes, you need to do that uh like send it around. You should be checking wherever that thing is right, but if you're offline you're on an airplane- and you know for a couple hours you're, you know writing stuff.

A

But if you really really need it to be like exactly this, then you should go full o cap.

E

A

But now you're incurring liveness requirements.

D

Three or four candidate systems with their requirements and test whatever potential uh version of usain location models on those, so they're blockchaining locations is one one example.

D

um We have like scheduling and running some land up in a you know: planner worker style site. um We have a function, call within say a rus program, that's running in one of those in that environment.

F

B

F

Mean like russ calling is basically just see it at the point. So.

D

Yeah, it just uses an.

F

Example of like a specific, concrete language function, call that you should paint.

D

In a traditional language um object that you should be able to define like this, it's not just kind of the implication of a whole program. It's rather there's a running program.

D

You want to send an implication, a specific function of a running.

G

I

D

Some function and location to some other from one program to another.

D

The definition of that object that you're sending the message like the actual function when one program is sending a message to another program, to run something that the structure.

F

Of the message is documented.

F

A doctor is that something like you send the doctor file that.

B

In my mind, that's all kind of the next level up of like a little b. So you want to get to one point about like multiple functionals operating on um a single piece of state in sequence or like. I think those are all analogous to conductor compose or a data processing pipeline.

F

Okay, sorry, I just didn't write it down.

I

It sounds really similar about mobizen consoles, like in our network and code, and then do some preparation for the result and also regarding them that we have a street resource handling, institute resource restrictions. Also, I would say that we need to take more attention on.

I

Should we run high rises.

D

Location stuff that lets you run the mixed processes, program, containers and other kind of v8, vms and so on, and so it just needs to extend the invocation model with for further definition like arguments so that you can do that.

D

So, if you're going to invoke a.

D

Unix program in a running unix system, you need to be able to define the operating system you're going to run and the program, so you need to sort of like bundle all of that into some individual. So maybe the evolution has to be kind of.

D

Parametrizable, where you can keep adding arguments to extend the application, so you can define things like the os the program, the primary server.

B

There's the hybridized view two ways of doing this: you can either we can buy all your programs uh or you can like literally implement the linux journal yeah. That is the better way of doing that.

E

B

Like you see one one, three from highly programmed.

B

You want to make like x86 in doing awesome. You can.

D

B

D

That's a way of getting portability or follow things. There's.

F

A browser example.

D

With it, I think it feels really fascinating. So for a lot of things that gets you guessing it's worse, it's better approach and then later you can make your nicer and more expensive to make. This is massive.

D

I

One, what I remember called they tried to.

E

I

With some engineering.

I

Yeah it was like someone died already. It was sold by his very small business.

I

All right, oh yeah,.

D

Yeah, I think we can just wait and the world will produce this.

F

Wow, I will post links their fonts.

I

The previous, like version of this course uh line, was like or not lesson in this. You have like some doubts regarding, because you know, like uh almost everyone wants to run javascript code. That'll skip code like as a function and trying to do stuff like is easy as possible, because we have like uh so russ is not so easy, but which is the javascript, and the javascript is much more widespread than rust and influence. We have some.

I

We had some soul that probably like, uh instead of using only restaurant and we could use something like v8, so it's engine from from, because we both run with assembly and both front landscape programs, and probably it's big because now, like you know, javascript can be compiled.

I

The only way to do that is just to compile the interpreter of javascript in the robot.

I

And I also know that guys, probably also stumbled across a problem and from other projects, also think.

E

I

How to be done, but for no business solutions.

C

Good, I mean I agree with everything you said, except for that last statement. There is an easy solution. It's not performing, but it's easy right. That's what we say. It's ready.

B

I think this is no assuming this is going to be successful.

C

I agree and like do we want to carry javascript into 2070 all right.

A

B

What do we encourage sensitive faults.

D

um I'm currently running gcc in a vm.

G

Which is pretty great, it's very slow.

A

Yeah, okay, but.

E

A

Also have to get into the whole conversation about seos.

D

A very easy way to go into all this is just using neighbors. They have the company locations, they have their own version of keynote, which is endo, which has like all the hearts and stuff. They have um capabilities that can be passed around through programs. The problem is that it's javascript everywhere, not just like at the top layer. It's like permeating the entire side. Certainly, what was the uh just using so the the stack has for each a ton of these problems. They've already solved these problems and made a solution.

D

um The problem is that it is a non-confident address, so that was a thing they sort of missed and b. It challenged every day, as opposed to just.

A

So the third solution is to use some dispute, so some escape.

I

Is like one which inspired by typescript and it could be about assembly, and it intended.

A

You know popular things.

A

Right, it's like at that.

E

A

Do we just rely on containers or do we also include the jbl right right, like there's a spectrum.

F

um Like is it the substrate web assembly, or is it something else? um I would almost say that like club assembly is probably the best looking one for um so we want to move forward on something that's very important too.

I

Probably the most the most be concerned about foreign.

I

Was invented for browser, uh and they thought about credibility like this is the first uh citizens of those and also.

E

I

uh Communication about this instructions that shouldn't be so like so large because, like the transferring data between I like internet, is matter and probably like the biggest concern here, is that it would be more optional to execute it on server side if assembly was initially intended was for server side and clients.

I

B

That is a more optimizable problem. Yeah.

F

I think we can. We can use web assembly just like as as the base and then, if we want to do more complex things, we can like access calls or whatever do something something subject.

I

Because, for example, they have proposals for uh not compiles functions like an enquiry and same way and stuff like that, not comparable assembly but use some stuff and then like lower some uh compiler characteristics and then like uses like more direct uh functionality, recommendation from runtime.

I

So here it looks like they try to push those.

E

I

So you know like there's a doctor, and the doctor allows you to like run programs or several programs. Like certain environments, like uh you know, run databases, application databases, but there is a like one idea that probably there are some use cases where you run only one single program in the document, because it needs to be uh secured into decent.

I

My personal desire is just to have a form of brutal machine that would like could be used instead of docker in several cases when you need to run one single program, because.

E

I

I

I

Small require a lot of kernel objects sheet groups stuff like that require somewhere in the kernel. Space and rob assembly will be the most optimal way to run the processes so probably going to be designed to.

B

That closes out on very sensitive time scales. Hypervisor racing is going.

C

Long term, like just so, we can understand like why slowly be okay, because.

D

I think we need to run.

E

D

Bunch of each other, we know we need to run all these other targets, and so we need to have.

G

D

That enables that, whether it's by doing like a something as horrible as emulating an x86 machine, wrapping that thing awesome or doing some other senior version of that.

F

D

F

Almost say arm emulation, I'm sure. Okay, sorry.

D

I'm just just some way of like uh running some kernel, as is on watson um or compiling a kernel through.

D

The as we just need some way of like knowing.

E

D

Those problems are going to be there and factoring them into invocations, and so on, so that we know how to so. When we design all these components. We know what that layer and that system has to be able to do so.

F

It would be good to.

E

Like see if we can take.

D

Some existing hypervisor thing um that is meant to run lips and and yeah figure out a way of running a docker container on docker on linux, on watson, and if you can do that, like you can do anything else,.

B

So just playing it's actually pretty fast, um but he does it by not emulating the fly transparently.

D

A

So, like is the concern more that are specific use cases that we want to support, or is it that we don't want to close doors behind us so that, as we discover more requirements,.

D

So, like people want to be able to run, you want to invoke some large pipeline program that already exists, and it's not going to be changed to do some processing as part of like a scientific notebook thing, and you want to be able to like content and address all of them and be able to reproduce the build of that, and you want to be able to like run that program. Take the outworld store. The apples continue to be being able to rerun the program to produce exact, same controls and so on.

D

You won't have a whole like interactive notebook thing that you can like free somewhere and then invoke reload show on the page about all the verdicts again and that should just work.

B

I wasn't subscribed underneath since, like you're not performing sensitive, historically yeah yeah. Most of these things are like we're not moving the cobalt highways.

D

Yeah in that example, yes, there are other examples where you are performing sensitive like you want to run some massive data pipeline around the um powerpoint yeah, and you want to do that like as close to the metal as possible and so you're going to write professionally.

D

F

So, even in those cases.

D

People are you're not going to re-teach the entire data science community to like learn some new things or write a new thing. In fact, they have to write in front when you think they won't use it right, so they want to use the existing things. However, you can compile them, so you can ask them to like follow a step to get the thing running, but it really they need some obstruction.

D

That looks just like whatever else they're using today, which tends to be containers.

F

And linux and uh javascript yeah all right in that community, specifically like tensors and.

C

F

B

And then also, there are specific situations where yeah.

A

Today, there's no need of an awesome way of doing this, but there's nothing like us. This game was a popular way of doing things. We can.

E

A

Now and do an emulation whatever and then later maybe there's a native velociraptor exactly from that user, isn't changing yeah.

D

That's more efficient and so on. You just need some underlying hypervisor layer that allows you to have these other runtimes yeah.

B

To implement that yeah, so just implementing some layer there.

D

Will be really useful to get going and that works better approach, but the other part is by having those candidates and targeted. You make sure that all the other things that we're doing will work. So you don't want to get like years into it and you realize. Oh, the application should have this brand, and now we can't have it.

D

In fact, we should probably self-describe the implication so that we can change the versions. We've learned that lesson way too many times.

B

Oh, so we have um we're adding uh the variation in container conversion numbers uh so basically deploy your active version. It's like correct.

B

But you have a version for the application for sure uh it's it's just methods, so we we send you whatever you want, so so the invocation data. So when you have an.

D

Actor calling another actor structure there if that actor location is like cross blocking, that's gonna have to get serialized out to something right now we just send hardware.

B

Number uh right now yeah, but how do you define a method number all right, so that is from.

B

E

D

B

It's weird because watsons don't have method numbers, we have the numbers um uh with their their multiples. Just right now, one more process both met, the name hashtag, the number kind of theory it does. Other proposals are like well, we should have a dynamic dispatch with the number um where, like you said, one or something like that with zero and then, like you, put the method inside the objective in such a way.

D

But I think what I'm just doing this out of conversation, we're not going to write in a location that is perfectly tuned for all systems. Yeah, you want that in location format. That means that's the self-describing that lets you. How do you move.

F

D

Yeah, the one problem with the ip names is that they attempt to be here like they can be mapped with big things and so multiple.

D

E

D

Like um if it's a big name, then it should also include it should include a bunch of underlying things in there we'll end up generating on the order of five to ten different things. And you don't want to. You want to be one of all of those under idealized.

F

F

Yeah yeah, I personally think this is like really.

B

D

Very spicy table but like it feels like products.

A

D

D

I want to have a web.

B

Assembly module within these other assembly. um That really looks like so we can use the component composers but like that here across things, um I really want to make this smaller. So people happier.

B

um So, like part of this project would be like trying to figure out how to do that now in a good way. uh The other part is just having addresses the addressing proposal is just like or the current ones. Basically, you have your reports and finally that's it so especially there. um uh So you say this guy. You can import these functions in these modules, uh but that that doesn't it helps with like sort of cross-column like they're, separate, fully separate.

B

uh They want their separate memory and stuff like that. It doesn't help when you want, like you just like you, have your set of hybrid parasites, uh bosom for everybody modules and they have some shared code underneath, but they actually want to share memories, called differences.

B

One is like library, level, uh imports and library level linking uh whereas terms like I, I import associated with functionality.

B

The other part here maybe I need to dispose of this topic, but it's like it's probably like we have a very large boss module and they have many of them and they all share a lot of code. um Ideally, I would be open to take that code out. I know there are tools that sort of try to do this. They don't do very.

F

Yeah uh by wasn't cloud, um they do this sort of thing where they just load things and components, and you can have the components, library and pull it and pile it together.

B

Can I so like what I actually want to do here? Well, yes, I can do that.

F

That compiles another actor, so it pulls in the different modules links them all together.

B

You'll completely do that one time so that, like when.

E

You deploy, you would say.

B

I have my last module and then add compiled type of change. It would bundle it all together, but my user is not.

B

B

There's a ton duplication, so you notice that, like I compress things and do whatever.

I

I

There are two issues.

B

Here one is the sort of like bigger macro issue with components. Basically, but like I have my module, I import other modules. I can either do this and then either at one time I can call it these or compile them together performance. That's one side of this.

E

The other side.

B

Is like, I want to be able to reuse.

A

B

Code, this is so a box is useful. It's also like for any type, because the thing is really useful, like I want to update some muscle control some new code, I don't want to have a person to read. The entire language includes, like a bunch of.

B

Else, yeah, but that also means that they can't run about opposition quickly and it limits that's kind of what gets you about conversations well because, like I write like that, they actually write after I compile it. So it also it means, and it means your target is no longer losing your targets.

B

D

In it and then the very first time you see a new program, you compound locally yeah, into a washing module, yeah.

B

It's worse, I I'm just certainly source. um You also follow like.

B

B

So, like I'm just concerned like the the time cost here gets really expensive.

E

It's also a source.

B

Code tends to be fairly large.

B

um It's not just.

B

E

B

Something like my own weekend, maybe like 20 seconds 2095, something like that.

B

G

What's your stack and then.

F

Okay, so something that like okay, so you have the bootstrap actor that you can compile and then you have an output of a huge wasn't binary, but you don't actually have to like. Have that as the actor that you have as an output, you could have that at run time, because you know the components that are going into it. So if you have a component actor, that's just a component or a library for other actors. That's a split line right, so you have like small, h, libraries that fit inside the size so before the actor.

B

F

Runs it just pulls in the cids and then yes, it's something the.

B

Problem, though, is like uh it's like that that works. It already has something like this. The problem.

E

Is quite often like.

B

Especially with the process like that, you build everything in one big bundle because of optimizations and the type system itself kind of requires you to do this. So one way to do this is like, if you pre-file a bus either there's an r um which is unstable right now. But if you do that, then you can laserly link on change, but that's again very rough specific.

B

What I would like to be able to do here is like take somebody's already like already bundled up, because that's generally programmed and split into small components and it's painful yeah, um but these wasn't the reason I couldn't believe this. It wasn't super super structured uh so like I should be able to do. This obviously do content chunking on awesome code. You know already our tools to do this and call them as a split. It will take the watson program and then profile.

B

It finds a way to like reuse, but like it figures out which parts are common, so you don't kind of put extracts common library uh and then, when there should be updates. I just like point to common library.

D

Okay, you could look at the entire stimulating yeah.

B

We do this around control blocks. I think you get some really nice properties out of that.

B

Like just looking at the binary itself, uh it can possibly make changes some offset somewhere or something like that. So you have to do this right. You have to look at the structure of the program in some way. Yeah.

D

B

So that's! These are the two things here uh for linking like. So this is your technology matches people chunking? uh So if you want to display well, you might need to be linking them because, like being like defensive offsets or something like that, uh where you need something like meta, impulsive format that lets you like uh your middle.

E

B

Modular format: uh it's not the final version, but let's do like different, like sub-plus modules and then pull them together with the second set of functions, but not full was much of themselves shards, uh where, like right now, we've kind of expected everyone's minds and top thing well. This is.

F

Can get into like sorry, I didn't I'm kind of taking over.

F

Right, it's you just apply deltas onto it's, you you! It's get that you just have delta, so you have a change in.

E

F

Code, but if we have it like a binary thing, you just apply the delta onto what you have. So the only thing extra you have to do is the delta.

B

Yeah we could just do binary patches and that would actually be interesting. Does that.

F

Actually, give you.

B

Attention to like whatever byte format, means trucking for certain stuff um where, like, if you have, if this bike already had a constantly binary patch, um then you'd be able to apply that if you have some kind of bike or something that wasn't suspended. But these are going to do this, like literally, do a program.

G

G

Now you're, like you know, I realized that, like this exported function is used in a couple of places. So let me like grab the exported and unexported function and make it a jar and then let that get like dynamically. You know link it under the.

B

G

By doing binary patches to like shove, it in the other places or.

E

G

Make an exported function right.

F

No yeah um yeah, I that's like the easiest way is where you have like the programmer do a bit of things to make it easier to chunk, but we're assuming that we just want general programs that are huge, that we need to shop and the people that have huge programs are most likely. The ones who don't want to change anything um just to get code running.

B

But like so, you could maybe get something out of the tools all right or something.

G

For gx or walls and modules right so so.

E

D

E

Is it's not just the.

D

Writing time, brick and pal time.

D

A lot of people are: writing different things, no matter how they're packaging and they're producing helpless, and today watson causes you to end up with these large big modules, and so it's even saying they could be exporting functions there. But you end up with different vlogs that you want to.

G

Reduce everything is the cost, though right you have. You are writing actors you're, written you've, written multiple actors that are using code? You are the developer, you know how to do this and if somebody else is paying to help put on chain a thing you're like. Oh, it cost me more. If I choose to not use steven's library or I have to.

D

Yeah so we're saying: how can we make that easy to do yes,.

B

So one of the problems here is like everything for people who very much.

A

Module that's getting uploaded right so like we want to bring things up around is fully.

E

A

The realm of authorization right, but.

E

A

The a lot of tooling will probably smart enough to do separate modules, but you know, as a weekend project, I decided to write my own language and then start writing.

E

A

And upload that right.

E

A

You know it's not really practical. You know real world use case. There's like those kinds of things right. It's like it's so deep in the system. That's not going to come out with an interface.

G

It sounds like you could apply this to like any programming language ever like any binary.

D

Yeah, but all of these systems don't have the benefit of content addressing right and where they do have it is it having that layer below where the file system might be like dfs, for example, might be um duplicating things underneath, so you might have multiple binaries and if they happen to align and they happen, they don't happen to have different like methods and so on and offsets, and they happen.

E

D

The same then, those minorities will do it, and so, when stephen is saying hey given that I found an addressing given that we know how much of these sections will be the same, given that in the wasm world, you know.

E

D

Massive programs- all bundled in one big thing, and we know that we're already gonna have much implication there. What can we do ahead in the post program like in a compilation layer to try and prepare these pieces for much easier duplication across the board, which is different from specifically externalizing some library and publishing as a package, it's rather saying that.

E

Library may be a package.

D

All together and be published, but there's a different step here, which is once you take that library in python or russ or whatever, and bundle it into into a wasm module. You want to be able to duplicate processes, so one solution here.

B

Would be potentially like when you basically do the wrestling team you might be able to try to keep things somewhat separate. You could also potentially, like maybe look like the each media representation inside.

D

E

Likely running.

D

Into this kind of problem with embedded devices, especially so it may be worth starting in a washing support. Well, yeah working room specifically.

F

D

Standards so that all the compiler tooling, that already exists, starts producing these outputs. I, like.

E

H

D

Of shards of curling charger yeah.

H

F

Like we don't have to do too much if we just wanted to be done about it like we could just say: oh we're just going to optimize the output web assembly and not bother with any language specific things um because, like we don't make any assumptions if we just have a general purpose, it just won't be as effective.

B

Locate things, but that gets harder.

C

Do you have in your head a target for the metrics you want to hit like.

E

B

C

Performance spin up performance.

B

I mean I want to be able to.

B

About a megabyte, stripped and everything um but identified like both actors and then compress them together, the compressed thing was basically the size of the original one. If.

E

I compress multiple.

B

Versions from build options.

B

It's not just deduplication it's yeah, okay, uh but yeah. The other is just compression uh and binaries, and all that kind of stuff yeah will give you that it doesn't.

C

B

Run time stuff so.

C

That would be that's the other part of this really.

E

B

A run to multiplication, so you can have like I.

E

Really want to be.

B

Able to have thousands and thousands of small little modules running, there is kind.

D

Of an interesting problem here like if you you have a a large content, address set of files growing and you want to add another file being able to smartly look at what you already have yeah, that that would be pretty.

B

Useful beyond the outside of this well, so the queer deployment model on top is actually uh users will create. The custom. Deploy uh will have nice defaults, but they can choose whatever they want and supply actors can also have its mistake. That which means like, if you have your deployment, you can play everything to play actor. You.

E

Can remember everything.

B

Like previously you'd be able to give it a different compressed thing, whatever you want, you would then disrupt the actual chain. So you can.

B

It's still somewhat tricky because, like we're going to charge you how much state you store, because people.

C

B

To get gifts and stuff like that.

D

B

D

So we talked about a lot of things. um We are all interested in different parts of this.

D

uh I think what we should do is list out um the set of projects that you would, you think, are needed to make these things and then, after we kind of make a pass of that, I want to go through and gather names of people interested in working on that. So, and I know that there's like many ways to cut this cake right so like uh let's try and kind of use whatever share that we're doing built up in the last day.

D

I can kick off below.

B

The computation side of things we definitely want to talk about the capability, modeling stuff and specifically the scene.

I

B

Happy vm context or anybody else, who's. Directing that context. I want.

E

B

How we get you hands layered there, so that we can work out that I also love to volunteer just to work on the capability reconciliation for.

F

Cool um I want to do like the dynamic linking and stuff.

F

D

And just if you just write something.

E

D

If you don't see it.

D

Concrete projects and then I signed up.

D

C

C

C

A

G

B

C

Stephen, do you uh do you think this stuff down here captures you the six and seven on there or should we have our own work item? I had it, I'm like looking at the rest of the oil section.

C

Look, I don't see the idle stuff here. No.

B

Pulse load yeah.

B

B

B

F

F

F

F

D

There's an attempt from mark miller and wonder from martin from ethereum oracle to define wasm specs in captain proto. Yes,.

I

So yeah the story was that so the entire specification was a little too slow in getting.

E

I

Last time try to draw language to themselves and close them with and then like previous year before they decided to move fix as a wheat and now he's abandoned. And now linux became okay. So it was.

I

E

E

E

D

D

Okay, how much does anybody need more time.

D

So we go through and mark who's interested in what is a very large. This became a task list more than a project was so let's try and use the first and if we need the second layer of the of the tree to try and signal interest if we can land it on the first layer that would be best because then the first um tier like if I got an invocation, normal execution capabilities, I believe wasn't linking equilibrium and so on can just be the products.

D

um If we need to go deeper, we can. That seems like much more complex should be starting to.

D

Yeah, does that seem like a sound way to do it, or do you take looking at these? These aren't problems that you.

E

D

Expect the same set of people working, you would expect within like say the red level.

F

Oh, I think it's mixed um there's there's some that is like it's specific to the single one like uh community outreach, probably includes multiple different teams.

F

Maybe not, or at least sure people to talk to the people that are on the outreach.

B

Team yeah, I think.

F

Everything except.

B

For coverage seems like a cohesive project, though so, like hypervisor is also called stuff, like you're.

E

B

All these, and sometimes you can't just take one piece- yeah, the communication model capabilities capabilities seems like it's, not that's going to leak everywhere, but it would be.

E

Good to have some feedback.

D

Yeah is really kind of part of the model fish.

E

The excessive, through the.

D

Hyperbacker too, but it's sort of part of the image model. um Let's just put it.

D

Yeah, that might.

D

B

F

Like the dynamic thinking.

E

F

Probably have some similar things could be one similar api, or at least I.

B

F

Right they're, different yeah, but communication is probably good.

D

Yes, do that capture your corner right, linking and calling the dimension any location.

B

Yeah, that's the way I see each one of these builds off the other yep, um where.

F

B

Thing is that you can start the innovation. You can't start eating this without waiting on the others uh by just saying whatever I don't care, it's like doing. Whatever works, fascinating, yeah, that's it, but.

D

This is like a good gap, but it's not a small part.

D

D

What do you think of where this is.

A

A

D

um Could we get some names of people interested in each one of these blue top level ones? It was like write your animal, although I guess we probably would want some number of people between two and six for these things, because if it gets too big, probably some things.

E

D

Defer to more groups, but I think if my working group becomes significant, it's still just kind of slow. um That's.

E

D

Do people agree, or is it really fast.

D

uh So here could we get like? Could we arrive at some number between two and six for names for the blue things.

A

A

C

Who wants a new community candidate.

D

Yeah, I might say that we have an ipvm, that's what's wondering it's like. Do we just try and have like a fairly lightweight.

B

Sink at a high level, maybe on a quarterly cadence.

D

Yeah, um I think I think, we'll probably get significant.

E

D

Progress by by having some kind of like planning.

D

Is probably reasonable, but then you would want teams that form to be working together very closely.

E

D

Certain things like frequently like chat, channel yeah.

E

D

With um people that are getting significant practices over time for duration of time- um and I do think we'll- probably get a lot of sessions like this- where we get what we have not much work and thought, and then we talk together in person about a certain thing, so we might think of rescheduling, sometimes in the future.

D

A long um one thing I will do later today is plenty like programming and playing um events for the enterprise community, and we use that to then live like oh. It would be great to have an fbvm sync at this point in time.

D

D

It's a lot of these outputs would then be taken by cov, so you might work a lot on the hypervisor side or on top of the hypervisor, so that once you have this tooling, how do you use it too? Then we haven't covered like things like scheduler, like writing, class scheduler and like doing locations on a network, and so like actually run a computational network with a scheduler built. On top of all of this, I think that's more where the cd work fits.

D

um So I would say the cld working group, probably like yeah is, is a user.

E

It's a user consumer and provides.

D

Feedback to um again.

E

D

Are the prototypes.

C

I think the biggest thing is: if we get them all involved early weighing in, they will be eventually happier consumers of it, because.

D

D

E

D

Windows and you have no, you cannot do containers and you have to run on a particular lab windows, machine directly connected to the lab uh instrument and you're a host. There's no virtualization possible, uh like that was a.

I

D

Important bit to absolutely consume like a uh make sure that they can actually solve the problem on.

A

Windows, ce yeah yeah.

A

So how do we get the groups talking about something like normal, where there's updates and ways.

C

I mean we have a weekly meeting right. We when we feel something is moderately baked that they can read.

E

F

Don't think it's much more and this microphone.

C

But I I don't think it's much more than what we have here in human consumable format. We say: hey we're thinking about a thing.

D

A structure with um with regular async updates. I have found the externally coordinated lots of groups, because um people also like to read the things they care about and they might care about it at different times. So they might not think they care about something so event happens. But if there's an async thing they can read later, then they can page in and then get on the train. um And so, if you come up with some cadence, that's good and natural for each team but they're sort of like reporting out summaries of.

D

What's going on that will that would be really useful to everybody else, um and I think that kate sort of depends on what the team wants to do um and has that went forward and so on.

D

Because for some teams it will even make sense to report out on like every two or four weeks for other teams. It might be like yeah every quarter or something like that. So.

B

Only two categories have only one.

G

Volunteer um we're pretty.

B

Good shape, this would be probably the fact that we had another session starting.

C

One third, uh I actually don't know what people's some people's handles are.

C

C

Dvd, can you give me your email just go drop into the google doc and attach it. You know.

B

Teddy's at the top.

F

We have two sections in the bottom at the top.

C

C

F

Oh, I feel like we did a lot of things.

D

uh Should we come up with a what do we want to kind of like leave it to each group and team to figure out what their next steps are, but should we set up like a cadence in which this whole verb? It's called the indian resonance just set the real presence yeah.

D

What do we think is a reasonable evidence like fast enough that, like we are making progress but slow enough that, like we, have something to report or something to discuss.

D

E

D

Six weeks yeah so should we have a six week days and have like a six week sink of I'm gonna, give you like a some virtual workout or something like that.

F

All right, this isn't any relationships. How often does the full of assembly people meet just how often people meet.

D

Yeah, do we want a weekly thing like that it could be a lot.

B

G

B

We'll have a good sense of that.

D

Higher dpms to problem here um for for some of you, somebody seems like really.

E

D

From tvs, yes,.

B

One part of my concern going into this is like a little bit of coordination, which is what these two deal is. We have a really nice foundation here.

E

um But balancing.

B

This for a lot of times, this is a new project. How is that.

D

All right we'll get a little bit more chance later today to distill outputs for the rest of the group, um and so we can figure out some of the cadence of timing and we're not there.

D

My sense is that we will, when we kind of talk about events in general, we'll find.

B

D

Spots where the community feels like we want together at some point, do a community time like that smart.

D