 So, so here we have a high silicon D02. Yeah Hello everybody. So who are you? Yeah, I'm Justin Zhao from high silicones And who are you from Lenaro? So you've been working on the D01s a little bit Yeah, I've got a bunch of D01s and stuff in my build farm for GCC development like ten of them Yeah, we have ten of them basically in our build and automation system And actually I use them pretty much all day every day So what do you think about this one? Oh, I'm classically been waiting for this for months since I heard it was in progress It was fun to actually play with one yesterday This is our latest border with our latest SOC on the border SOC name is P660 and each processor has 16 cores in it and now we have already enabled the XG, the SFP, the SATA's and the PCI-R So there's like 12 SATA ports? Yeah, there are eight SATA ports for one one processor And there are another spare four SATA ports for another Another processor. There can be two processors. Yeah, now we just enable one processor, but maybe a month later Maybe two or three months later. We will enable two processors on these boards and each processor will have 64 will have 32 cores in it and with 32 MBs or 3 cache in it So we're talking 32 cores of ARM Cortex 857? Yeah, yeah, of course. That's a lot of performance, no a little bit more than a tablet Okay, yeah, obviously. Well, this is more of a server enterprise class machine and this is a glorified telephone This is designed especially for some heavy loader scenario So just when I use my D0-1s for instance, I build Duke a lot of continuous building test integration of the Canuteo chain I destroy most hardware. So amazingly all my D0-1s are still running. All of them are still running I can't talk about some of the brands I've killed in the last few months, but yeah, my tears are one still run We're not using all 16 cores because it doesn't work yet But the four that are there work reliably, and that's more important. Which brands have you burned up? I'm not gonna mention that I kill hardware pretty religious pretty frequently The latest kernel 3.19 has been enabled just as you see Okay, so what kind of you can see the There are 16 CPU cores in one one processors Yeah, and yesterday when I was doing a total chain builds and stuff It was nice to monitor and actually see that I was using 16 cores and watching my processes migrate based on load That was kind of nice Okay, so it actually worked. So you say was the first day you saw it? Yeah. Yeah, so what happened? Well, of course, I tried to build GCC on it And the one to 16 64 bits Is not available yet. Yeah, then our problem was the network around this went down It had nothing to do with the hardware that kind of screwed the rest of the test Well, we will try to resolve it. Maybe a week later. We will enable the Ubuntu 64 bits just just weeks, I think what does that change when you have the 64-bit Ubuntu? Firstly we are supported more More application based on 64 bits secondly, we Of course the performance so we are aware will be better I think and there's a lot of things that actually need 64 bit once you start getting in the cloud computing and running open Stack and all that you kind of need for 64 bit top the bottom. Yeah. Yeah, now I just This this morning by this money. I just tried to enable the Linux container in our in our boats. Just I can show it I have stopped it Yeah Never did So what kind of questions would you ask him If you were to try to ask questions about the status of the software and the board and everything we have also deploy or deploy our Our Lava server on our balls. Oh nice. Yeah, you maybe I can straight to you Okay, I see the We just This this several we are in in clothes My circle. Oh, sorry. I'm sorry. I don't connect to the network This is why we love demos I Just a close the laptop What are you doing right now on your on your tablet? Oh, I'm debugging a GCC problem So remotely accessing some actually logged into an arm V8 board sitting in Cambridge So this is the lava This is not in lava. This is in the so we have Two different build farms within Lenaro, we have the lava lab and then the toolchain team has its own lab We have sort of different uses of that. We just had a meeting about integrating what we're doing, but We beat on hardware very differently Whereas a lot of the existing lava stuff bases on building kernels and rebooting them constantly We have a tendency to bring up targets and let them run for months and months and just do continuous compiler testing every time We do a commit the GCC we fire up a build of however many Configurations we support the stuff and then run full testing and then validation on every commit It's pretty hard on hardware because we often see if you little hit, you know 100% and keep it there for a long time. That's why I kill boards. I make them overheat really good So what do you need to run all that stuff on the on boards? Well, there's nothing like running on real hardware to make sure your compiler is producing the right code And then some of the stuff we're about to start using lava for is actually using our we just got out of the meeting for it A few minutes ago is we'll be actually using all of our kernel our toolchain builds from head of trunk to build kernels And then make sure that it actually boots in the hardware because we want to catch Compiler bugs that create kernel bugs before any of the distributions do because right now there's a sort of a lag time Between them getting our compilers and getting back to us about obscure bugs So we're just trying to have better and better better validation of the whole ecosystem between the kernel and the toolchain Because that's kind of what we know. I mean, it's kind of our service. We provided the community is validation on all this stuff so making sure there are no bugs in In what area of Linux and where well, I mean so as far as a toolchain is concerned is you want to compiling the kernels easy It's a matter of does it still boot and you'd be surprised how many times for a long time The kernel development was very stuck on specific Compiler versions often not upgrading for like a really really really long time because toolchain changes freak out kernel developers because we make their life difficult Lenaro is kind of different because we're really having to push both the compiler performance and the kernel performance And so unlike a lot of people I'm in Lenaro We typically run at the tip of mainstream for both toolchains and kernels And that's kind of our ideas testing both and you kind of have to do that because you know People that are building on V8 hardware want all the performance improvements that are we're working on pretty much all the time So you were at high silicon, right? Yeah. Yeah, are you as I need at Lenaro or yeah, maybe from this year I will be a tiny Fox on Fox on the virtualization field because we think as a virtualization is a very very important function for for some special for for the network server or some Some some seem like this always pretty big in network. Yeah. Yeah, and I think you're completely right Virtualization is the big thing these days for cloud and this kind of class of hardware should support that rather well Yeah, so this morning. I just try to Enable the container in container on our boats firstly and now maybe I can show it to you I I have created Container named just in container and I have studied and Then I can I can attach You can see you we have In entered a container it is not In the host now, you can see the name the name is just in container not is Lenaro developer. Yeah, do you see it? Okay? This is the container. Well, I just enabled it this morning and there is Lava website we enable it a lot of Service service in our boats and you can see it from this website Yeah, yeah, this means we will enable a lot of love by our ourselves arms arms inside servers and machines, okay and We also do some like Do some testing and benchmark on it, but The data is I think is a data Can't be public now. So I can why Secret performance no Because we think the data is not a precise so far so maybe we will we are to perfect and improve the benchmark and And then we will give out the Data to So what's your brand new hardware? Yeah, it's not quite all there because we just enable this border by last Saturday So we're just glad so many many works needed to Yeah, so what do you think about the work they're doing at high silicon? Well, of course, we're glad to say it and stuff So I mean it said I've got a lot of Rmv7 hardware specifically and so my these are the ones have been pretty reliable And so I'm been kind of looking forward since now that Lenaro is Shifting a lot of energy into the Rmv8 since the Rmv7 is pretty solid and stable We actually need some good stable Rmv8 platforms No has been working on Rmv8 for three plus years or yeah pretty much and a lot of that's really coming to fruition But the problem is we actually haven't seen much Rmv8 hardware and what little stuff comes out It's typically very very early silicon and very very early hardware and so we kill it in about a week makes good door stops But they said our diesel ones have been good, so we've been hoping that the diesel two is going to be a good work horse and Soon later we will provide a more hardware to you so that you can work better on this This device, okay soon No more than three months, I think no more than three months. Yeah, yeah, yeah, there'll be large quantities for all the developers At that time we will how now for hardware for all our development and For all our developer and for our customers and it's something how much performance do you think there is in there? So Good question because I'm less I mean from my side of the world. I'm less worried about performance Because I'm only compiling the compiler what I'm worried about is stability on their heavy load And so in a sense compiling a toll chain will keep all 16 cores busy and typically keep them busy at a pretty high CPU load So I'm really curious about it Just playing running loads loads for a really long time And that's how I've had really good luck with the diesel ones a lot of boards I'll build two or three toll chains in a course of a few hours and they overheat and stop working So it'll build step faster than what you used to right that would be nice, too Because we do so many back ports and validation testing the faster it gets done the more validation I can do basically and then in the faster validation happens It means the more and more so if we want to start doing things like using our toll chain Building the kernel and then testing on the hardware right now We just don't have enough hardware and time in the day to do quite that exhaustively And as we get more and more automated It's a lot easier for us to just keep it busy all the time And before there was a bunch of people around the table here. You were showing some stuff about it, right? You were showing demos and what were you showing people about this board on the screen before yeah? I just I just Show you should there we enable our lava server on our arm inside both and and we also enable I mean it container just by this this this morning and we can We have already enabled the multi boot mode for example the PXC the provisions on FS and the SATA now maybe we can show it to you. Okay. Okay Reboot that's what you guys are working on when I was earlier tonight is PXC. Yeah. Yeah. Yeah, we have already Already we can start it from satan now and also enable the pixie and an FS and something like this now I can just show the the Sata boot putting up This is Here here we can select is the one is the fall putting up from satan and the tweezers putting up from from PXC and for is putting up for by provision command now I I can select the one For demo here we can select different route FS for example a mini system or open source or Obamac maybe we can try the obamac The company the currently begin satan enabled and Network is enabled The Eric Eric's network is not available yet Because I needed to rebuild the the kernel for for the X container. So you see you've been talking. Yeah We use that for benchmarking. That's good. Yeah. Yeah now we have enter the Obamac system, okay So that's cool. Okay, so That's a pretty cool board PCI express. Yeah, this is a tool to say this is slots for for this one processor and this this tool is spared for for this one Maybe how many run for per What is this? For processor. Yeah, they are there for gigabytes around how many maybe there are I don't have a maximum Memory is how how many but the another at least it will support one one two eight Gigabytes, I think yeah, it should be good enough. Okay Johnson yeah, yeah, they are a to satan interface for for some of them To connect this one process and there are two or there are four another Sutter ports for this one for yeah for other call and there are two SFP plus interface for network and there are two normal gigabytes It's a net port for for this one Yeah, yeah, so thanks a lot for showing this off and talking about it. Thank you. Thank you very much Okay, thank you very much