 This is our friend James from AWS for those of you who weren't in the cloud track earlier. He's going to talk a little bit more about how AWS can help Debian and perhaps maybe expand a little more about what you talked earlier about for those who weren't sure. So hello everyone, I'm James and there are precisely two slides for this. This is supposed to be an informal meeting, so let's be informal. I've been looking at ways to see how Amazon can further help Debian and I guess one of the first things that I've done since I've gotten to the ability to do this was to get the AMIs up for EC2. And then beyond that, you may have remembered we had a release recently and so around that I think it was the weekend before, I thought, well, we should really see if we can get the CloudFront CDN up and running to be able to assist this. And that was part of my talk this morning, showing how you can access cloudfront.debian.net and you can get the CD images through that and you can get your regular archive access through that. And so this became a, hey, let's throw this open to the floor and see what else we can do. Now at the moment, Lucas also has a project he's been running where he's been recompiling binaries on EC2. Now that could be just testing. I can't remember the rest of the details around it. Those binaries might not actually end up in the distribution, but at least a vast amount of testing can be done. But we can do lots of stuff. And so I was after ideas, requests, any kind of information anybody wants. Let me, I'm pretty much going to throw this open to the floor right now. So let's have some ideas. Who's got an idea of something they want to do and use a vast amount of computing storage to do something interesting with? Come on. Okay. If you're still, if you're still, come first, you guys, come further forward, please. This is a small group and it looks like an empty room on the cameras right now. So please, come right here. I don't. Yes, please. Now, let's turn that one on. Okay. There we go. Okay. I'm managing PyCould and OpenCL packages which deal with GPU computing. So I think I could, I'm not using auto testing while building those packages because usually Debian machines don't have any GPUs on them. So would it be possible, for example, to use Amazon for testing them? Yes. So this way I know whether this is coincidence that packages work on my machine or only whether they work on more machines. Perfect. So we have one instance type which has GPUs on it which you would be able to test on. Here is $1,000 to test it with. Wow. Next person. Any other questions now? So that credit is valid for a year and that's a brilliant use of using the AWS resources. There's one instance type which has GPUs at the moment, they're the NVIDIA Tesla GPUs. There will be some more coming at some stage soon, so more modern GPUs on the instances. So can I use that for harvesting bitcoins? It had to be a S, it had to be a S. Give me three of those please. Yes, but I was going to say which country you're in because depending on which country you reside in, I think they're now illegal in some countries like Vietnam. So yes. Anyone else got anything else interesting they're thinking of doing or would like to try? Surely I'm greasing the wheels here, who's got something? I have actually personally no use but I could imagine that we could build some images for Debian and for Debian Meet which is doing some medical or some bioinformatics calculations and running gene sequencing programs and so on. That's something actually I'm really interested in doing because I've got a bunch of customers who have sequences and some of these sequences have little computers in them. And some of these little computers have plugins which instead of storing all of the sequence data locally, they throw it into S3 because of the storage. So we have large numbers of sequences, throwing that at us. So what should they analyze that with? Well, I think they should analyze it with Debian Meet. So yeah, one of the things I've been thinking of is, and this was part of the discussion we've just come out of, was what other images should we master and have ready to run on the cloud? And this is not an Amazon thing but this is all our clouds. We currently have a base image and customers, users of those images can actually then instruct those servers when they start to go and install the Debian Meet or Debian side blend or whatever you want. So should we have them prepared and ready to go without having that extra step? We can automate all of this. This is just scripting. But that's an interesting one for us to think about. Yeah, if you'd like that, we can do that. We should talk about this. Are there any other requirements? I mean, what would you need? I don't understand gene sequencing that much or any of the others. Me too. I also bundled the packages. Andres, you are exalted as the expert around the world. I know. People think I'm the expert, but I'm just building the packages and I trust on the users. And if they report this or that is wrong, I think it's wrong. But it is not so. I'm happy with this. But yeah, we should ask our users to use Amazon cloud to do their gene sequencing. I think that's one of the advantages that we've got. This is a pretty awesome operating system that we've managed to cultivate here over the last 19.99 years. And I think a lot of the packages that people in the scientific community and med communities and others, we have packages which a lot of others don't. I can't even give this one away. So GPU testing. I think I was a brilliant one. I really do. Because yes, that's a piece of hardware that's sometimes a little bit difficult to get hold of to do that testing on, especially if you wanted to do it at scale. Analytics. Any other storage? I had Zach come to me and say, what about CDN for www.debian.org? So that's one of the things that we could do to try and reduce some of the load on the current website and obviously distribute that globally. Hosting snapshots.debian.org. Anyone familiar with snapshots? Yep. And the bigger archive.debian.org? Yes? That's a question. If you might host snapshots on an Amazon cloud and your credits are running out. So you are bound to the Amazon cloud and trust some snapshots? So for those initiatives, I'll probably take a different route and not be using the credits. So credits are just one option that I've got. I've got to speak with other teams to work out longer term things for bigger things. Have we thought about QA? Anything to do with QA? Tell me about it. I don't know. I just know that our customers DevTest is one of their big things is launching test tests on cloud. I don't really know how our QA team works. Well, my understanding is that's some of what Lucas is doing now, but it's only some of what Debian does. I thought he was doing more of the build cycle, but it'd be nice if, especially server related packages could be tested in the cloud. Not even server related. Not even porting packages. So, unfortunately, Wookie's not here. But one of the things I was thinking about was what if we've had an ARM64 emulator and we wanted to rebuild every package? How many packages have we got? 37? 39,000. 39,000, something like that? This isn't something you're going to do on one machine or two machines. This is something you're going to do on 8,000 machines for four hours or something like that. So these are things that I think would be useful to do. We are all going to benefit from that. And it's probably been brought up a hundred times already, but just for one-off development, you know, that's what got me into Amazon Cloud was, you know, I've got crappy DSL at my house. I need to pull down massive amounts of data, do something with it, maybe upload it somewhere else. Yeah. Just access to maybe, I know you've got these credits now, but in a general sense, if a Debian developer wants to hop on and do something Debian related. Contact me. Is that the best way to do it now? Yeah. Yeah. That's pretty much it. And I want to see Amazon continue to assist this community, because it is a very important community to me personally, but I think to Amazon's customers as well. And I think you'd probably say the same thing about GCE. Because we basically want to start a war between you guys so that everything we do becomes free. We're resisting that completely. We're mostly saying positive things to each other at the whole conference and supporting each other. If you want blood, you've got to go to the other top room. Anything else? I mean, we're going to end up with a pretty short session here. Has anybody else in the room got ideas of things they want to do or get Amazon to do to help Debian? Test upstart versus system D performance. See, that was what Steve came and said right after my talk. Decided instead of Leonard and everyone in that session. So what do we think? Should we do system D and just not worry about what's happening up there? Let's make that decision here and go for it for all our cloud images. We could define 100 processes for them to manage and launch them both in both of your clouds and test the performance of your clouds at the same time. Well, actually, that's already come up. Yes, so on a serious note, this is what Steve Lancaster came up and said, look, can we make a system D image and a standard image and test them and let other people test them, not just Debian, but let our users test them and experiment with them. It's just such a hard problem because there's so many variables as far as hot bloke stuff. So we should find those issues. We should fix them and we should fix them on mass. We shouldn't be fixing them one at a time. We should have, you know, almost a crowdsourced effort to find all those bugs and get them reported. Anything else? Yes, please, microphone. I have not slightly related questions, but are there command line tools in Debian main for Amazon or not yet? Not yet, not currently. I think we're probably going to come up with the same thing on this. Do you know, Mike? I think there are out of date... Yeah. The eucalyptus tools have out of date versions, which may not work in the ideal ways for the current Amazon cloud. I think that's the situation. There's a bunch of third-party utility. So I don't believe the Amazon produced SDKs, so for Python, obviously, that includes Boto, are installed. They're not packaged currently. This is something that Boto is, and I don't think it's quite as old as you guys think it is. I'm going to have to double-check with that. Charles Plessy maintains it, which we're in contact with. Unfortunately, when they moved me from the release team to the sales side, I've kind of lost track of a little bit, but he updated it from what I understand before the Weezy release. The point is actually really interesting, though. The rate of which the improvements of the environment is happening now, one of those improvements is that the current round of SDKs support the idea of getting time-based security tokens for secure access to APIs. And I think this is something that Mandy was talking about this morning. Is that similar? The normal way to access most Google APIs, including the cloud one, is similar to the OS 2 thing. Amazon's adding that, you're saying? And I don't think that the tools or SDKs that are currently in main, or even in unstable or testing, support that. Unstable, they should, because I believe Charles pushed 3.0 there recently. If not, he's going to, because there was a Python dependency that was missing, that's been packaged. But then we're talking Python. And unfortunately, we've also got Java, Node. Yeah, there's six or eight of them. And we're also talking unstable, not stable. Yeah, so we've got another two years to go. What was the question? I was asking specifically about the command-line tools, because I know that Botto is quite recent. It's 9.2 something, so it already includes... Is that in stable? No, it's in unstable. Stable contains rather old version of Botto. Unstable contains version of Botto with Glacier support, for example, which is important to me. But I was asking about command-line tools, to manage instances from... The basic stuff you can do with the Eucatools list there. Yeah, I mean our command-line tools, there's two of them actually. The older command-line tools are Java-based. I don't believe they're packaged, and I can't remember the license off the top of my head. I do know that our newer unified command-line tools, which are hosted on GitHub currently, it's an open-source project. It's run by Mitch, who did Botto. That isn't currently packaged, but I would love it to be packaged, because you just want these tools to work. Now, they might not be installed by default, and that's a separate argument that we've just had, and I think it's going to percolate for some time whether or not we install tools by default into Cloud AMIs. But into base, Debian, this is... I think I can't remember what the license is. It is definitely DFSG-compatible, and I'm sure GS Utility is probably also similar. We would love these to be in main. We would love them to be available to everybody. The same concerns come up over the lifecycle and timing of this, but an older version is better than no version. You should put those over to our team, actually. We'd be happy to look at those. I didn't even know they existed. I know about the Java ones. The AWS CLI one is under heavy, heavy development right now, and I've got a... I have it running on this other alternate operating system that's on here, and it's just Python, and it's nice. It's all JSON goodness and loosely copying and all that. So come on, some more ideas. What else do we want Amazon to do for Debian? Does it use Boto, by the way? I assume it does if it's Mitch. Yes, it is a new version of Boto. It's pretty up-to-date version of Boto. I think he's a little bit here. It's very good. It's very good. It's just hard to package his stuff sometimes. He's good. He's just like any other developer. Latest and greatest doesn't give a crap about how hard it is to actually push it onto a machine sometimes. Well, that's bleeding edge software, isn't it? Please, someone else. So from a QA point of view, as we've talked about, I'm interested in anything else that we can do to help. I don't think he's either here, is he? I don't know. But any kind of automation that we can do stuff at scale, let me know. Cool. Should we go up and watch the discussions in the next room? I don't see why not. Excellent. Please help yourself to t-shirts because there's no way I'm carrying these back to Australia. So for this room, for each, and we'll just pass them around and whatever. Cool. Thank you, everyone.