Unimatrix Members Call, 14 Aug 2019

Previous Meeting

⏯

youtube image

►

From YouTube: Unimatrix members conference call 2019-08-14

Description

https://gitlab.com/unimatrix/meetings/-/blob/master/20190814_mom.md

A

Okay, I think we're recording now. Okay, so I see some other people joined.

B

A

B

Hello, how are.

C

You good thank.

B

D

Hello, it's sibin from costine hi steven welcome. Thank you.

E

F

So on this side we have also I'm mr villain.

G

I really hi everyone and paris here for the first time of all these meetings, can you hear.

A

Us well yeah, very clearly, okay, good! Thank you! Okay. I can see some other people. Can you present yourself.

D

Yeah we we've also got uh patrick joining in from costine. One of my colleagues here he's joining in for the first time.

A

A

A

Okay and uh where's uh you from.

A

H

C

H

Sorry, everyone it's from vision, hello,.

A

H

I'm good just my app was blocked so.

C

I

C

J

Oh, I am otavio campana, it's the first time that I joined.

A

J

Are you well tired because it's vacation now so I have to take both the children it's easier to be at the office.

A

Yeah, of course, okay, thank you for joining it's nice to hear from you again.

J

C

Hello guys this is hill from davao, hello, hello, you! How are you I'm fine? Thank you.

I

Hi this is, this is patrick from custom.

A

Hello, patrick.

C

A

Thank you for joining um okay, so um I guess we'll start with the agenda, um so we have a draft agenda. Do you have anything to to add to that.

J

A

But we can't see the agenda on the screen: okay,.

J

A

So what can you see exactly on the screen now.

C

I see nothing, you can't see anything. Okay, yeah! You can see okay, yeah good. Thank you.

A

So is there any other topics, you'd like to add to the.

A

A

Okay, then we'll we'll we'll go with this agenda for now and you can uh you can have things as we go on so the first. uh The first topic is the naming of the the project, so we had a poll that that really launched uh two weeks ago.

F

Yes and uh it actually, uh most votes went to uni, matrix and sopsa with a draw between those two, and so we started.

H

F

Matrix as a product name, so unless anybody disagrees, uh we continue use that as a product name and if we need to change later on that's possible. If we need to have something, that's able to register as a trademark or something, but for now we I think it's good to use. This is a product name.

A

Yes, so we actually registered the name on gitlab, so we decided on using git lab over over github, because git lab is kind of it's more independent github is owned by microsoft, and uh gitlab also has a little bit more.

A

Features when it comes to devops and things like that, so it seemed like a nice platform. So I went ahead and registered the name there and so far we have two projects. There is one is presentations, the other one is meetings, so I just added there the the last meeting agenda. We have some more meeting agendas to add there, but uh and minutes and things so uh I will also I will send like an invitation to everyone to to join this group.

A

It's a so it's a group called unimatrix and under the this group name, you can add projects so so right now we have uh only these products and we'll I'll add some more products to build the reference platform as we go.

A

Are there any comments on.

A

A

So I I'll send them. You will basically get an email invitation. If you have not already registered to git lab, you have a chance to register your user there and then you can join the project. So it's an open, open group. Anybody can actually add themselves and then we have sub projects under the group. So that's how it's organized right now.

A

And probably I I don't know if you know anything about gitlab, but they have different price models so right now we have like using a free, it doesn't cost anything. But probably we want to use a bit more advanced model, so there will be like a monthly cost that we need funding.

A

So that's something to discuss in hong kong. I guess.

G

I mean up to now: it's I think. It's quite few costs we're having this zoom account and we.

G

Because the whole idea is only to cover the exact cost, we have for tools and like to.

G

A

Okay, so I will, I will probably send out the invitations today, so you can. You can register your user and add yourself to the project and then you can check the the minutes of meeting and so on.

A

All right, if there's no comments on that part.

A

So the next point is the development platform.

A

So previous meeting we decided to use uh the raspberry pi 4.

A

um This platform is quite new, so I had problem finding any good distribution that we can use, because I want to use the 64-bit architecture and if you take something like raspbian or debian, or something like that, it's all of them is built for 32 bits. So I actually built from scratch my own distribution using open embedded. So I will my my idea is to put the the things you need to build um on this gitlab project. It's a couple of open, embedded layers that you need so.

J

So let me understand: do you plan to to open source the layers.

A

J

Oh, that's really. Nice.

A

Yes, so you will be able to build everything uh so um also I played around a little bit with the different layers, there's a couple of layers for building docker and things like that. But I don't know if you know about this meta virtualization, which contains a lot of virtualization, also container stuff, but it's it's not really well maintained in my in my eyes, I don't know if you have any better idea.

J

Well, the other possibility is to work with build root which has support for uh for docker engine and so on.

J

J

I know I don't know what you think that could be better.

A

My idea was basically to take the stuff, we need from meta virtualization and put it in our own layer, and then we just build what we need and another idea is to use something like baleno os, but they have some. They have some players, but then they depend, I think they have like we.

A

They depend on docker, for example. You need to build dockers, some on your own and put on your, so they don't really solve the the whole whole use case. We need.

J

I think we could try and work with the octo, and uh should we find this problem, we could try to fall back on build roots.

A

Yeah, okay, so I will make it go for using meta virtualization, but probably it's going to build a lot of things we don't actually need so, which is? I guess, it's okay, we can optimize. You.

A

A

So anyway, um I hope I can do that before I'm going on a conference next week, so I hope I can get that out on the open source on the gitlab this week.

A

Are there any comments on the the development platform anymore, any other ideas on how which how we can build stuff, or are you fine, with using yokoto.

B

A

Okay, so um I hope to get that done these weeks and then you can probably next conference call you. You can have a chance to build stuff before that.

A

So the next item is the steering group and charter yeah.

G

So the idea is that we said we will be starting a steering group but a sort of a governance group in the same outline that linux foundation proposes and we have drafted a charter which we have sent off for legal review, and I expect that to get that. Like a week before we meet in hong kong and then when we meet in hong kong, we will review that, and hopefully we can all agree at that point and then send it to the linux foundation for approval.

G

It's it's real, relatively basic. It doesn't describe that much in detail, so I think the the. uh If we get a steering group or governance group together, we can assign some other groups within the within the project to produce a little bit more detailed specifications. You will be achieving.

A

Do you think we should put it on the git lab site before hong kong meeting, so people have a chance to we.

G

Can but I would I would like just waiting for to receive it from from uh legal reviewing and once I get it back from legal, then we can put it on the gitlab. So.

A

On the git lab, we have two two repositories now, so one is private, so that one will be only for uh people that are have joined the project and the other one is open. So I guess we can.

G

A

On the private one, yes.

G

So I've agreed with the with the it's actually lindelt that we use for onward as well, so they're quite familiar with this couple of things, and I've distracted them to have this done by minimum a week in advance because we probably need some iterations before we agree with. That is that okay.

J

uh Yes, so pair uh just to have an idea. uh What is the effort required for people who are taking part to the steering group.

G

I don't think we.

J

G

That in yet really, but the idea is that the steering group or the governor's group should consider consists of members that actually contribute so to earn earn a seat in the in the in the steering group. You need to be contributing member.

J

Okay, but what I'm saying is the the following, so we have davide that is uh taking part to the to the contributing uh sessions.

J

So if you look for people taking part to the steering group, I may be interested, but I really would like one to have an idea on how much the effort is expected to be because before offering, I want to be able to be sure that I can I I can do what we promise.

G

Yes, I do understand that then, and but I think it may be a little bit too early yet to say exactly what is required. We will. We will try to.

A

Detail that, but I mean it's more than taking part in the telcos, so it's actually contributing development resources to contribute on the project. So so that would be, you know uh contributing on the source code and on like the container layers and so on or with budget, I guess with the contributing for the the running of gitlab and so on.

G

But I think I think what you mean octavia is that how much will be required for you to participate in the.

J

Yes, exactly, that's that's the question.

G

Yeah, I guess maybe now in the in the startup phase, uh as the proposed charter is laid out, where a little bit of more technical detail will be sort of described by the engineers is for for the steering group to review that and approve that to to proceed.

G

So, uh but let me think about it. I I can't really say if it's a few hours a month or if it's, uh if it's enough to have a child, go every two weeks in initially or if it needs needed to be an hour every week or something I don't know. I.

J

Know? Okay, I was more than that. I was scared about the fact that how often are we expected to meet face to face because for me last year, is good twice a year could be acceptable more than twice a year in this phase of my life is tough, so that was my biggest concern.

G

Yeah well, the the thing is that we, when we started to discuss this, we did not foresee a an organizer similar to like onvif. We more based this on the concept of an open source meeting where people keep continuous contact through web conferences and that type of media.

G

So we did not anticipate to have that many physical meetings actually uh and for now we have. We have proposed meetings in conjunction with the on with meetings, and I can't really see if we should continue to do that if it doesn't cause a problem for any members or not. I don't know.

A

I think another idea is to co-host with linux foundation. So when there is like a conference like open source summit or embedded linux, uh we we could have like uh meetings on those locations.

G

Yeah, maybe this is typically also a task that we have to to.

G

Us and agree on: how often do we actually do need to meet physically.

J

So, okay, so the message is I I I we as tech have I've been taking part very often to uh to the activities that this week is on vacation as everybody else in my company, so not a problem, uh for he will follow the technical discussion and if you need people on the technical committee I may I'm on the understeering platform. Excuse me and stealing group then I I may offer my my time if it is not that much.

G

Are you planning to go to uh the almost meeting in hong kong.

J

No hong kong, no I I'm I'm a.

H

J

G

So uh that could be an option then, if, if.

G

People come in these meetings. Maybe we can meet them.

G

How this develops and what how much contribution is required from from it from a steering committee point of view. Okay,.

G

Good, so uh I will post the uh the charter once I receive it from legal review and then we can look at it and then we discuss what we meet in hong kong or for those that do not participate, can feedback to us, and we will bring that into the discussion in hong kong.

E

Okay, can you expect next week or next week after that or something oops? Where are we now, in date,.

G

Next week,.

G

A

So next call is on the 28th.

G

Yeah before before the 28th, hopefully I can actually circle back to the indulge and ask them where they are. With this, I don't.

H

Think I don't anticipate.

G

A

All right: are there any comments about the charter or organization of um steering committee and so on.

E

Not from us, I think, we'll see how it progresses.

E

H

We just want to be more clear what what kind of stuff we uh from from our side from hydrogen inside we can, we can contribute, or we can do or we need to.

H

We need to go multiply when one will be in hong kong or.

A

A

At this point uh I mean we're first, we want to like launch this, so we have a common platform which we can start with, which is this raspberry plus like a minimal firmware that has a docker engine and the basic things you need to do wrong. Containers plus uh I mean I, I guess the next step for me would be to start to discuss.

A

uh You know what common container layers we want to maintain. So I see like stuff like tensorflow lite. You want to have a version that runs on the platform, we're choosing what other, what other things, the apis, that we need to get video into the container to to do inference from the container and produce some kind of metadata or events or whatever from a container.

A

So those details with is um my idea about what we need to discuss in in hong kong on the technical side, the basic apis to get. You know, a video analytics application running in the container to produce some kind of metadata.

J

Frederick, yes, I think we should start uh a little bit uh earlier. I mean first before doing what you say. I think we should clearly state what our device model is, because the when we were discussing in one week there was an implicit model that was uh defined by the by that standard. But here we are not we.

J

G

J

Giving only for granted so we need to clearly state what our model is and how it is expected to work at.

E

A

Well, I see the model, it's not it's very different from one, because this is not really a device client kind of model at all. So this could be warning. I mean this container can be running in in the cloud or it could be running in a device, but yeah I mean we already lost we already discussed about. We need a use case, um some use cases which.

A

You know it's where we should focus so we need I. I agree. We need a more high level discussion first and then maybe go into technical.

J

Yes, also because uh uh the the way I I interpret uh your the proposed uni metrics is not only on the camera side, but also on the server software side. So we implicitly make some assumptions. uh The software runs on the cloud. It's a local bms that somehow orchestrates the algorithms whatever. So I think we must give at least an overview of how we think the whole the from a high level point of view how the whole uni metric is going to work.

A

A

So I think uh we should start with that in in hong kong, the the high-level use cases and high-level.

E

A

I guess we're sliding into the.

B

A

H

The agenda for.

A

The hong kong meeting, so we already have used cases.

C

C

E

A

Yeah the charger- I guess I think, last time we said we're, gonna have two separate tracks, one will be technical and one will be more uh governance.

A

So um from the high question side, maybe we needed to meet two separate meeting rooms during uh the hong kong meeting.

H

uh Yeah, I will check in the I don't know: um when can we confirm all the participant for for one.

A

Commenting yeah, I don't know, we need some kind of registration for the meeting, so somebody has to look into that this you can do self-registration or you can just send an.

G

Outlook invite.

F

A

Okay, so so we'll we'll send an invite on outlook for, for the hong kong me and everybody who will attend will accept.

G

A

G

Maybe we might consider to uh have possibilities to connect via zoom if you can't be in hong kong, if you want to listen in.

J

Yes, I'd like to yeah.

F

For the session.

A

So maybe first we needed like a comment session because I think some some discussions, maybe which all of us should like the the use cases and things like that. And then maybe we can separate for the technical stuff and the.

A

Governance so yeah, I think we should. We should put together a more detailed agenda before the next telco.

A

A

And then we will send out this outlook invitation. So we know how many people are.

A

A

Are there any comments on the on the agenda or the the meeting in hong kong.

B

So uh christian is a uh my question. Is uh uh how long would we take a meeting.

A

I think we planned for a full day, so I think adler you planned for the whole day right from nine to five.

H

Yeah yeah, that's true uh mask is just one interesting question regarding to only for meeting do you guys receive any some traffic alert based on current situation over there.

F

I read that some countries have issued a lot like us, australia, so.

G

We yeah, we are governed by the uh the the foreign office here in axis the swedish foreign office, and there are no restrictions just to avoid meeting places. That's what what it says.

G

Yeah it's the same for the uk, that's the advice of uk as well yeah and julie uh from one with from gi. She has been in contact with the with the hotel and they they run different type of conferences constantly and they are capable of of providing all the services in-house.

G

So if it's a little bit of an issue outside the hotel, we have the possibilities to stay inside. They they say it's safe. They don't see any concern. Of course there might be a little bit biased because they hold the booking, but that's at least what they said. So I guess it could be transportation problems. It.

F

Could be a transportation.

G

But so far it uh I mean the the airport is an issue, but uh I think reroute replaced on nearby. I.

H

Heard I I've heard a lot of flights I cancelled recently and yeah. I don't know, I don't know if, if that will cause any potential issue for all the participants coming from every everywhere from every country- and I I don't know.

A

I probably should uh take some have some extra time for for for some consequences: yeah yeah, so we we're traveling in a bit earlier.

A

You know: okay.

A

Okay, um I think I think that's all we had for today. Actually so we have some action points on exercise for next time, so we'll uh we'll I'll send out the invitations to join this.

A

This group- and hopefully I can get this layer, so you can actually build something and then we'll send out the outlook invitation to join the hong kong meeting, we'll prepare the agenda for the hong kong meeting. So we can discuss that more in detail next time and.

G

A

G

Put the charger yeah the charger on.

A

G

As soon as we receive it back.

A

All right is there anything else that you'd like to discuss today, or we can end a bit early.

J

Frederick, if you have five minutes, we could talk a little bit uh so that I can fill the gap that I had from the previous uh meetings. So if you have a five minutes so that I can align myself with the current state at the end of the call, okay.

H

A

A

A

Okay, then, thank you very much for joining and we'll talk in two weeks.

E

C

Thank you, yeah. Okay, thank you. Thank you. Thank you. Bye thank.

B

C

G

Will you learn or will you dial in or.

J

No, no, I'm still here, I'm still here so just to understand a few things to so. You said you want to you to to to publish a couple of layers for yocto, which is totally fine for us as well. um The thing that I am not going to understand very well is what will be, let's say, the level of functions that will be provided by these two layers.

J

I mean: are you just going to uh export two layers so that you can compile docker inside or are you also thinking about uh publishing some source code, for example,.

J

Something that takes a snapshot from a camera connected to the pi and is then wrapped in some kind of process control so that it does not consume all the cpu and so on, and then it streams data back somehow.

J

So, what's the level of the detail you want to implement in the two layers that you are going to publish.

A

Well, first, uh it will be like just using whatever is there, but in the long term um we are considering to publish an inference engine to to kind of.

A

So you can run inference on uh like the tensorflow serving api, but interfacing whatever hardware you have underneath. So at this point we have the raspberry pi, but you could see it will be like an inference engine which is layered, so you can replace the lower part with another back end, if you're using let's say openvino or another machine learning engine okay, so that part will probably uh gonna open source from access side.

A

We also have other parts like our idea. Right now is to for you getting video into the container is to use um open opencv, but so we need like opencv, is a huge library, so we need probably to fork it or yes, so we use only part of it for uh the parts that we need, which is basically right. Now the capture capture api.

J

Okay, okay, and what about the client side, I mean this orchestration thing should be somehow managed by a bms.

J

A

A

It doesn't have to be vms, I mean it could be uh at this point we haven't decided on orchestration. So at this point for deployment of these containers, we suggest to start with only using container d, the continuities uh rpc grpc api for deployment, and then, if you want something fancy for orchestration, you can actually run on top of that. So you could, you could run kubernetes or docker compose or something in on top of that api. So the low level api that we want to focus on right now is only container d. Api.

J

Okay, okay, I I understand now on your goal.

A

Okay, but I mean in the long term of course, there's so many things, nice things with running kubernetes. So probably, but it's it's gonna be an optional. It's gonna, it's not gonna, be something that is required of every device, because kubernetes is so huge. So you know a lot of devices will never be able to to run that thing.

J

No, yes, but my vision is that um how to stay? Maybe you have different algorithms that are in different containers and based on allowance metadata. Whatever that you have in the vms, you want to turn a camera from a function into another one.

C

J

You should turn on and off, containers.

J

But I don't know if this is a too far vision.

A

I think it's good to have a vision, but for right now the api we're focusing are more of the low-level apis. So it's more! It's not really orchestrating at this point is like deploying and without containers. Yes, yes, so you.

I

A

You could actually run you, don't even have to support containers, I mean you, we had we, we used container as a as a format to package the thing, but you don't actually have to run run it as a container.

A

Okay, so I mean you could use any you don't have you could just unpack it and run it as whatever or you could use use a more sophisticated virtualization if you, if you want to do that, but the we don't put any requirements on how you run it.

J

Okay, because that was another question that was in my list, so you said you want to exploit raspberry pi 4, because it is a 64 bit okay, but there is still a plenty of things that is shipped with 32 bits. So I was scared if you, what was the level of the target thing?

J

uh Okay, another question: uh this is more technical, so container the uh container. The. If I correctly remember it's written in go, am I correct? Yes, so uh go is compile the language right?

J

Yes, so eventually, what are the dependencies at runtime?

J

Because I'm wondering if uh how is this thing going to increase the size of the whole system? Based on your experience, once you compile container? Is it like a regular binary, or do you have to have additional libraries or environments whatever.

A

Well, normally, go is always compiled statically. So it's all that's why the binary is quite big. So if you take you run c, for example, it's you know: 10 megabytes, container d, this that's probably around 20 25 megabytes, so it's quite, and so they don't use the uh a common run time at least on arm. I think there's it's possible to do that on on intel, but on arm. Basically, they don't share any common libraries. Normally you can it's compiled into static binaries or everything.

J

Okay, oh wow, 20 megabytes, it's a lot for our binary, it's quite big! Yes, yes, I'm asking, because I had a bad experience a few months ago, I I was trying to do some experiment on my own, and I was thinking about compiling a javascript engine like spider monkey on for cameras just to propose eventually, eventually a possible common language because javascript, you know you can compile c to run on top of that, so it could have been very, very nice also for performance.

J

So I started trying to compile v8 from google, which is really really ugly. uh It's not something that you can just download the source code and compile it. It's really a mess, because you need to download from google the tools to download the source, and then it does a lot of tricks. So everything it's everything but linear for compiling and the other one that I checked was a spider monkey and this spider monkey was a half c and half thrust and the rust compiler has so many restrictions. So that was also very painful.

J

So when I saw go, I I was a little bit upset about the number of compatible.

A

Platforms, yeah I mean go, is it's it's made for cloud from the start, so it's still, but I think it's still getting smaller. So I mean this is not really intended for embedded use from the start, but uh things are progressing.

A

uh I know it. I mean docker is working a lot from container d to move to because they also see this use case of embedded to make it to make it possible to run on on our type of devices.

A

So I think things are happening uh quite fast. Actually,.

J

Okay, I will study on that and let you have feedback.

A

There is a yeah I mean docker. Has they have two editions, there's the community edition, and then they have the enterprise yeah, so that one is more optimized for for some reason, so it's smaller yeah, then you pay a lot but.

J

Okay, okay, so okay, I will I will study. I will check a little bit things here and there and then I will provide you feedback. I sent already the invitation for accessing the unimatrix project, so I think please keep me in the loop.

J

I try to do my best to take part to to the calls, because I see a lot of opportunities in this idea and and not only for cameras.

A

That's great, that's right! Exactly we see a lot, the I mean the interest should be bigger than on with so I mean the intention is to have more uh different kind of companies involved in this also.

I

Well, from from castain's perspective, we were in parallel, our embedded team were investigating the use of docker containers on our embedded systems, and I just happened to mention mention this to jeff, and he said um he invited me to this group because of that.

A

That's great, so what are your? What are your experiences with the size of stuff.

I

So we we've it's still early days, um but I should point out that um yokta and open embedded do make a distinction between compile time dependencies and runtime dependencies and a lot of the the dependencies are compile time that so all the go libraries. So I I'm I'm not too concerned, but it is. It is worth having having that on the radar. So just in case.

A

Yeah, if your device is already challenged when it comes to flashing and ram, you know putting container d there is it's going to tip it over so yeah for our from our side. We are it's more for future products. I guess than because running I mean continuity is one thing, but if you want the whole docker demon and everything, that's it's quite huge and then on top of that, kubernetes is enormous.

J

Yes, we also have the same problem. Our current architecture will not support it, so I am pretty confident the future architectures will, but the current one that we have, in my opinion, is far too small yeah. uh Just to give you an uh last feedback, uh you know that we work with the octo. We have been discussing about that so many times in the evening drinking beers yeah. I am doing investigation nowadays on the build world, because bill root is faster, easier to use and.

J

Still we are, we are starting to think to switch from yorkto to build.

J

It's easier to expand.

A

Well, you're welcome to contribute on that. We, but just to get something started that we can you know, use together. I think we can just launch this uh open, embedded stuff to start with.

J

Yes and then I will see what's there and then we'll try to understand it if we can replicate it.

J

C

Okay, thank you. Thank you.

C

F