18693 >>: Okay. Go ahead. >> Dennis Gannon: Okay. This is the last wrap-up session for our meeting. What we've got up here is our primary program committee, less Judith, and she was giving somebody a tour of her office, I think, last we heard. So maybe she'll come in. We are not going to do death by PowerPoint at the end. So we'll spare you that. This is intended to be general open discussion. First of all, I want to thank everybody for coming here for this meeting. We are just thrilled by the turnout and the quality of the presentations. When a program committee takes abstracts and selects a program based on a few paragraphs of text, you don't know how it's going to come out. And I've just been thrilled by the quality of the presentations and the quality of the discussion that's followed. And so what I'd like to do is as soon as each of these other people make some final comments is to just open this for a discussion with the whole group about things we didn't do, things you wish we should have done. Your feelings about the future of the cloud, what you're concerned about, what you're excited about. I think that would be the most interesting way to conclude this. Oh, and also a general discussion about things like what we're going to do, what we should do next in terms of this meeting. But, Arkady, did you want to say anything? Whoops. We still have people coming in. >> Arkady Retik: Please come in. So I want to say that this event started as an MSR event for limited kind of group. We saw a lot of interest from people overseas, and maybe just to introduce myself, if you don't know, I'm responsible for worldwide faculty engagement. And since we heard a lot of people from overseas want to come, so we made an effort to many of you come and we're very happy that despite this short notice and a lot of arrangements, you came. We are very interested in academia collaboration, and it's important that not only connection be with us, but between you too. But also since it's just start, we're very interested in your feedback. So this is very important for us, because we're doing this for you. So any feedback, I think this was a survey hosted. There's some books with some feedback there. You can do it anonymously or you can send e-mail to any one of us and tell what you like, what you didn't like, what you want to hear from us. Any programs you're interested, like internal, any other things, you are very welcome. >> Kent Foster: Just a brief comment to what Arkady talked about. We actually put this together in a relatively short period of time. It actually reminded me of back in 2006, after Microsoft Research had their faculty summit, Jim Gray collared me and several colleagues and said let's put together an E-science workshop and let's do it in October. That was three months. We said well there won't be a time for a full call for papers but we can pull as many faculty as we knew who were interested in this topic and invite them to submit abstracts. And we ended up with a conference with 220 attendees from all over the world, which is now going to repeat in its fifth year this coming October. So I think there's a track record here of bringing together faculty who have an interest in a particular topic, get a vibrant discussion going like we've had the past couple of years and begin to form this into something where the discussion can continue going forward, perhaps with a formal call for papers and a program committee and a paper review and potentially academic journal publications, et cetera. We did something very similar, again, back in 2006 with the topic of game development, which evolved into a formal conference that is now co-branded by ACM four different Sigs and is now independent of Microsoft, and we think that's a great success as well. So I'm personally hoping this topic of cloud futures will have a similar trajectory. You'll go back to your campuses and talk to your colleagues and begin to talk about what next year might offer. It doesn't have to be here in Redmond, we've moved these conferences around the country. So it doesn't have to be here in Redmond, although this year it seems that co-locating this event with the external research faculty symposium gave many of you who don't have regular contact with people here at Microsoft Research an opportunity to hear what they're interested in, what they're funding, what kinds of discussions they're having, and perhaps deliver some value to you that you wouldn't have had if you just came here for two days for cloud futures. The other thing I'd just add to what Arkady was talking about, we're very segmented here at Microsoft, in case you haven't noticed. Arkady and I work over on we call it the evangelism side of the business where we're working with faculty really around what is being taught in the classroom, what you're teaching your students and how we can equip your students with the tools that they're going to need to be successful in the workforce. But then we have external research here at Building 99 that's very interested in getting to know all of you from a research perspective in developing research collaborations, and you heard this morning from Dan Reid about what XCG is doing and the hope that we have to build a very vibrant community around this cloud topic. So I think that we're working together. This is very much a collaboration between the group that Arkady and I are involved with and DPE and extreme computing group and extreme research as well as a collaboration with all of you, because you took time out of your busy schedules coming from a long distance in many cases and you're jet lagged and you're getting through all these sessions and you're all here at the wrap-up, which I'm just very impressed with. I just want to thank you all for your efforts and hope this conversation will continue. >>: To tack on to everyone else's comments to simply say thank you for your contributions. This has been great hearing all of the ideas, all the papers and the discussions. Our group is keenly interested in understanding what kind of research can we support on the cloud. Clearly this is going to be a dialogue that's going to unfold over the next several years. We want to be active in that with you. We are building websites and resources that we can offer to the community to help bootstrap your efforts, but we really need to have your input on what kind of services we should be offering and what kind of resources we can be offering in addition to educational resources to help start projects up on the cloud. So definitely we'd like to hear from you and for you to be actively involved and even go back to your own research institutions and think about helping articulate for your domain, for your communities, what is appropriate for the cloud today and what workloads simply can't be supported so we can have an open dialogue in the form of papers, blog postings, et cetera, both on our websites and your own sites as well. >> Dennis Gannon: Let me just add along those lines something that I mentioned when I gave a talk at the end of the faculty summit meeting, is -- and Dan alluded to this today, that we have engaged with a variety of organizations to provide Azure resources on a substantial scale. We've done this now with the U.S. National Science Foundation. As I mentioned the other day, we have just concluded a contract with the European Commission to provide cloud resources. Farbresio Gagliardi, I don't think he's here today, about you he's led that effort. And what will be happening with that is there will be an open call for people who would like to have some substantial or at least as much as we can afford access to Azure resources in Europe. In Asia I'm working on similar projects. I've learned that the U.S. was hard. Europe is five times more difficult. And Asia is five more times difficult than Europe. But we'll get there. And we will make these things, these resources available. So at this point I want to just sort of open it to general comments and questions and suggestions, things you think we missed that you think need to be brought out. This is your chance to stand on a soapbox, and tell us about it. Or just anything you want to talk about. >>: [indiscernible] technical University of methods. I would like to thank you for the great event. I think it has been very interesting. Something that I would have been very interested in hearing, mostly because of my bias, would be talks about the architecture on technical challenges, scientific challenges that have been overcome to make Azure possible. As a researcher in cloud, I need diversity in this kind of stuff. I understand that technical details would be not public, but I think many others will be and we can learn a lot of things with like papers from others like DFS and so on. And another thing that could be interesting, not for the event, but for Azure would be to have a kind of academic version where researchers can look into the architecture and can try to improve things like an academy person of Azure where we have the code buy level and we can improve things and do research on top of that. And finally to teach courses regarding cloud to students, it would be great to be able to apply for a client to get hours for the students, to use Azure for their [indiscernible] because if we have to pay or if they have to pay it becomes difficult to manage. Thank you. >> Dennis Gannon: So let me just comment on some of those points. One is that the lack of presentations about the Azure architecture. We've actually tried to keep the presentations focused on your work and not necessarily our internal work. But I think that's a great idea. A lot of the things we can talk about -- Dan this morning talked a little bit about the architecture of the datacenter. The architecture, and you've seen some sort of practical views. User level views of the Azure architecture and the preceding tutorial. And in terms of the details of the architecture, that's a more complicated problem. And it's largely due to the fact that Azure itself is evolving very, very rapidly. And I've had many faculty members come to me and they said: Is it all right I have an experiment I want to run I want to get in and modify the resource allocator in Azure. Unfortunately, we're using a commercial service here, and there are companies that are currently using Azure. For example, we have a chain of pizza parlors named Pizza Hut in this country which uses Azure, and I would hate to have to see the Pizza Hut people explain to their customers why the pizza is late because the resource allocator experiment failed. So that's why I've not had much success with doing it. I'd love to do it because we get a lot of requests from computer scientists to do that. I think as we look at the platform as it evolves, in the future who knows. Neither Roger nor I who work pretty closely with Azure have any indication that that's going to happen anytime soon. But they surprise us every week with new things that they do. Roger, is that? >> Roger Barga: Great. Spot on. >> Dennis Gannon: I think next time. Yeah. We should do more on detailed architecture. Good point. >>: We have architectures from research distributed community many would recognize those who are architectures at Azure we certainly could have them come and talk in general the decisions they made in crafting Azure and making the design decisions they made, what were the trade-offs, I think that would be interesting to the audience. We'll certainly fold that in. This year we were folding on the applications that sit on top. We certainly can drop down to that next level. We'll even extend an invitation next year also to Amazon and Google to ask them to also come and share their design decisions, because it's an interesting space, but this year we targeted a little uplevel. >> Dennis Gannon: That would be very interesting if Google would come and share their design decisions. >>: Jeff Dean has given some very candid talks. In many cases he shares secrets about how they made design decisions. If you look at his talks, they're fantastic. Lots of nuggets in his talks. >>: It's an event called Lattice with the system. And the most interesting thing was the second event where they invited people from [indiscernible] and the talks that they gave that were [indiscernible] still containing challenges on how they overcome them. But sometimes [indiscernible] which one was the problem that they were trying to solve. >>: Congratulations. This meeting has been really, really good. We've been able to meet several people from all over the world, touching on trying to put applications on the cloud. I just wanted to share this observation, especially with Professor Gannon was talking about the application going down. There is this pattern called the open source community. I do not know if you're aware, Hedge Pin, Yahoo! others are supporting the system development of packages which can possibly run a cloud. And that offers a lot of opportunities for us to play around, to understand what goes inside the cloud and how to research and contribute there. But a similar effort from the Microsoft site would be something really worthwhile and interesting to us. Because many times I've phone here and inquired and hit a wall. What goes inside Azure, we don't know. What is inside Azure, we see it but more at the user level. I think at the user level trying to cook up applications it's nice, but somewhere along the line as researchers we want to peek inside, under the hood. So your observations, please. >>: Well if you get your wishes realized there's going to be a lot of people wanting to run experiments and to run research and so on, and it doesn't seem right, indeed, to do this on the same platform as the one that is used to maintain the quality and the regularity of the pizza supply of the United States. So what about, I don't know, Research Azure, Academic Azure, someplace where we can indeed ask you to change the resource allocator or do crazy things, possibly even, I hate to mention it, for free. [laughter]. >> Dennis Gannon: We'd love to do that. It's just a matter of what do we have at our disposal. And at this point all we really had was Azure as it is. And so our group, Roger and I, run by Dan Reid, is literally buying service from our budget, from Azure, in order to give it away. And that's all we have. We can't even -- at this point -- I say at this point if we bought a container of servers, we couldn't even get the code ourselves to put on that system. I mean, it is evolving so rapidly, that we've looked into this. But -- but that doesn't rule it out in the future. I'd love to see it happen. >>: So you left me without a question. I'll formulate something else. First, thanks for this wonderful workshop. I think it would greatly help us if you could provide a list of Microsoft people working on Azure and exactly what they are doing on Azure, at least keywords so we can select whom to send questions to. >>: So we do have -- so that's probably not a great thing that we could do, but we do have -- if you go to our website that we offered Microsoft@ -researchmicrosoft.com/Azure. We have an engage alias right on the right side send us a question we have a team that will get answers to you if the questions turn into how did you implement X that would probably be a little more difficult -- I don't want to set expectations for that. But I'd also push back on I used to be a distributed systems guy when I was in graduate school. We had the Linux code, et cetera, et cetera, but at the end of the day the more interesting stuff stood on top of the operating system. It was out in the bowels we weren't compiling the operating system, there was a focus should we focus on on top and have transformational experiences or do we really want to be working down in the plumbing. I realize the audience will split if we split the room the system researchers over here are saying give me a RAC, give me a code. The other half I want the programming models I want the security, et cetera. So we are making some purposeful choices here on the audience. We're focusing both limitation-wise and I think where you might have the most impact. Because at the end of the day it's about what can people get done with the cloud, and it's the experience, not the flops, not how fast it goes, but that's just a personal choice at this point. >>: Another point of contact, if you missed, was an academic pilot. Azurepilot.com by [indiscernible] previous session. It's an excellent point. Get in touch and ask questions. It will be more oriented. >>: The areas of research. >>: We'll have all the presentations up for you to be able to watch and I'll send everybody an e-mail saying it's up here is the link share it with your colleagues. We realize with three different tracks it was impossible to catch every session. Other questions? Comments? >> Dennis Gannon: I have a question. This is for the group. I mean, one of the things that if you -- from Dan's talk this morning and the interest that Roger and I have and others within our group, that we consider to be really the exciting frontier on much of the distributed systems and applications side is really that connection between our client devices and the sensor networks, the data streams. These are, for us, the grand challenges and how do we connect. I know there's at least one iPad in the room. How do we connect devices like that to cloud applications? Because right now where we stand with the cloud is if you look, we're all surrounded by cloud applications right now. And they are on our cell phones. And we use them and every day to connect to some cloud service someplace. Well, that model of computation, that's a programming model. That's a systems model, is extremely rich and potentially very rewarding for innovation. And so that's what's driving our group to a large extent. And so that's where -and so my question is: So how many people are interested in this, say the client or the cloud application space among this group? Oh, actually quite a few. And the data streaming area. Yeah, as well. Paul talked about that. And we had several really great talks on that as well. So I think this is -- we would like to encourage people to look at that and we would like to see more creativity there. Because that's a wonderful area. We have a new person back here from us who as I said she was off in her office, I think, someplace. >> Judith Bishop: I was showing somebody around. >> Dennis Gannon: We heard that you were showing them your office. It must be a big one. It's taking a while. [laughter]. >>: We were just making some comments from an external research perspective, anything you want to say about the workshop? >> Judith Bishop: It's on? >>: Yes. >> Judith Bishop: I'm just bowled over by the fact that so many people came to this dream of Kent's, really, and that it was such a success. I was chairing two sessions, and I found the papers really, really interesting. I can't single out any one of them, but from my point of view, just listening to what's been said here, the topic that hasn't been raised is the idea of software development in the cloud. So who's interested in software development in the cloud? Okay. Great. That's something that I think could chew up quite a few cycles and also something that is due for cloud -- well, use of the cloud because it's not something that we do every day. So when we need to use it, we pay for it. When we don't need to use it then we don't pay for it. I think that's also a very good one. From the external research point of view, we haven't as yet put money into funding programs with cloud. But something obviously we'll have to start thinking about, given the interest you've all shown here, and thank you for prompting us on that. >> Dennis Gannon: So on that programming model issue, and program development in a cloud. I've been struck by the fact that we might be in a bit of a rut in our programming style and it would be very interesting to look at sort of new approaches. For example, if you think about these current client-to-cloud applications, we have typically a set of services in the cloud which are then mapped, mashed up really in a separate application. Is it possible, for example, to write a single application that really spans, goes from the client device into the cloud? You write one program and it resides somehow in both places. And the system, the compilers manage to distribute that application automatically. Is that conceivable? Is that a reasonable thing to do? >>: Yes, that's the holy grail that people have been looking for many, many years. >>: Holy grail. >>: In fact, we already have a prototype version of this running on top of ->> Dennis Gannon: We'll go to Barcelona to see it. [laughter]. >>: You are invited, of course. >> Dennis Gannon: Other topics that were not covered that you wish would have been covered? >>: That's one question. When you start to think of the cloud as where things are happening, where it's all going to head, how does it affect curricula? Krishna put up these topic of things, these are ways, things that people have to know to talk about what's going on in the cloud. How many people have students which are actually prepped for all of that? You know, because you just can't -- if you do it all in one class, take up your whole semester, right? We're talking that this is going to filter down and have impact on our curriculum. Arguably, all the way into the freshman year, right? And where? Our department struggles on how do we revolve to that? >>: [indiscernible]. >>: Yeah, actually there were two educational sessions, and there was -- very interesting development done by University of Austria University they proposed some proposal where you actually can integrate the topic, because there is no cloud applications. So they came up with this proposal of teaching it as part of a distributed system, part of OS. Part of application. And they actually, it's very good news. They actually developed the course and will make it available to all of us through Faculty Connection. It's Microsoft.com/faculty, and we will make it available to all of you. And there you go, you can start teaching in May, around May it will already available. >>: For a course, have a software engineering person, that's how we did it. We offered we thought we would have no one. We had 15 students. It was brilliant. We had projects all over. So we had students making projects in all of the known and available platforms at the time. It was very [indiscernible]. >>: Actually, Rosa reminded me of one application we are working on. And I forgot to mention it in my talk. [Chuckling] it's about the elasticity. So the dynamic adaptation of an application on different infrastructure. We are working on the importing mobile phone virtual machine on cloud. So trying to move the entire system. So the operating system together with the application and to a cloud in case of energy efficient problem or in case of need of how to say more computing power necessary to run application on the mobile phone cannot be done efficiently. So we are moving virtual machine between the cloud and the mobile system. >> Dennis Gannon: That's very interesting. >>: Could be interesting services because you can provide on a cloud the service for thousands of users. Okay. >> Dennis Gannon: So just one further comment on the education theme. One of the things that we have experienced at Microsoft, when we tried to hire people out of college and/or graduate school, and we ask them, all right, have you written any good cloud apps lately? Or have you done anything that is a robust distributed, scaleable service application? And there's very, very few that have had that experience. >>: Bill howl from the University of Washington this has been a fantastic event and I'm happy to be a part of it we talk about the importance of data but how we still talk about the cloud, still doesn't reflect that. We talk about it like it's a platform, there will be applications that run. The idea is that the applications are sort of siloed. But one thing that's nice about the cloud is that physically co-locating all this stuff, data and applications in the same place, solves a pretty big chunk of the problem getting them to talk to each other. Now with corporate customers, they don't actually want to talk to each other, maybe they want to put up walls but with science applications, having the climatologist put their data up with the same place as the biologist offers new opportunities that weren't possible before there was a cloud. The way I sort of think of the cloud is the database of everything. I'm sort of bias because I think of everything as a database. But I think maybe if we could, if we're offering suggestions for the next one, maybe to think about the cloud as one big database rather than a bunch of siloed applications would be something that I'd be interested in hearing about. >>: A challenge. Do a cloud challenge. Just come up with some application and have like they do it for the agents, you know, the bidding systems. And just do it for the cloud, just come up with a problem, put it out loud, distributed prize and have students develop across, and we could use that as a benchmark for cloud solutions. And then we'll distribute that and that could be used for teaching in the next semesters and things like that. >> Would that be something that people would both be interested in seeing at our events both demonstrations and maybe have people submit demos or a cloud challenge in advance and people here could actually vote on it make the decision here so actually make it more interactive to see what people are building and doing. Something we kicked around, we kicked around a demo or poster session; but we thought the timing was too short for that, but what would the interest in that be. Have a birds of a feather session where people whoa re writing code can interact with other people writing code on different clouds. What's the audience's thought about that? >>: I think demos are good if the students are here, because then they could have workshops and they could exchange code. Because they're the ones, have you ever tried building the applications for the clouds, they're really hard. And they require a lot of work. So the professors are not going to do that but the students will. So if we can just bring them and they will serve as replicators and we'll see. >>: So would there be value, see a show of hands, if we were to both have a demo session and also maybe do tutorials for students and have hands-on birds of a feather session so students can learn from each other go back to their universities and faculty as well. Show of hands people interested in something like that. Good. Thank you. Good suggestion. >>: What about a cloud summer school? Any interest? Okay. >>: Whose summer? >>: We talked about it. Trying to figure out, gee, can we do the workshop and then do a summer school? >>: Better locations than Redmond? >> Dennis Gannon: I think Judith is pointing out that ->>: You just have the things that were submitted, why don't you do that. Post-proceedings, chapter book style. >>: We like to publish. [laughter]. >> Dennis Gannon: So one thing about a cloud challenge, I like the idea of sort of picking a set of problem, challenge problems that people could work on. We do have the classic one for data centers has become kind of boring, it's tera sort challenge, how fast can you sort of tera list of items. It's a terabyte or -- whatever it is. Terabyte dataset. And I think we could do better than that. And so I challenge you to come up with good cloud challenges that we could get everybody interested in. >>: I think one of the possibilities we have running what is called Imagine Cup competition. Imagine Cup competition, everyone is interested know about. I'll speak to the team, I'll suggest maybe we'll come up with some challenge that faculties and students can work together and we can implement it. But for this it's important that we have start teaching and doing something. Okay. Great idea, thank you. >>: We do something with Imagine Cup faculty mentors are very important part of that. So we have Imagine Cup going in about 80 different countries. There's competition at the regional and national level and so forth. So if we can get a cloud category, cloud challenge category for Imagine Cup, that would be very good. Other ideas? Comments? >>: So I want to share with you a thought. So I think that four part [indiscernible] science, whatever you want to call it, it's a nice thing. But actually I think if we must start sharing data, because it's mostly this -- first of all, data and then you want to process them. We need stability. I mean, we need a stable place to up load data on the way or perhaps protocols to expose our data to the communities. And places where this data can be collected, shared with the academics. And actually I'm just facing this problem because we found a way to measure from power absorption algorithms and I want to start a community doing testing worldwide. And actually, okay, where can I start? And actually there's no answer to this. I mean, so I think we should also think not what, what we want to do with the cloud, but also how stable can be these initiatives over time? Otherwise, it seems to be a lot of wasted time. >>: I have one perhaps heretic thing I would love to see. Because I thought it was really great that you had Amazon and Google here. And I would personally learn a tremendous amount if I could see the same app deployed, that is, developed for Azure, for Google, for Amazon, and, fine, let whoever do it. It brings out the best of each of them. Just to see the differences and see what it takes, see how different aspects of the different platforms are utilized and taken advantage of and perhaps advantages or disadvantages that anyone will admit to. Of course, there are no disadvantages to Azure, of course not. But I would personally be just -- I would love to see that. I would learn tremendous amount about the cloud from seeing something like that. >>: I think that might carry a lot more credibility if academics would do it. We could consult to make sure. And I'm sure Amazon would do the same to make sure that our technologies are used in the best possible way. But again it would lend credibility, especially if you do it over a spectrum of applications because you would see where some excel and some fall down. We've done a little bit of internal analysis ourselves to see how we stack up. But it might be an interesting discussion to share amongst people who would be a good application suite and then go ahead and do the evaluation yourselves. We could consult. >> Dennis Gannon: So if we're finished, I have one final thing, and that is actually to really thank Kent Foster who is the guy who really pulled this all together more than any of us, in other words, he dragged us along. [Applause]. >>: Great. I think we're going to have buses leaving over the next 45 minutes from out here to take you back to the hotel at the Marriott. For those of you that are leaving, if any of you are leaving to go to the airport, you can see the receptionist up until 5:00, if you need help with airport transportation. One last show of hands, because I know everyone's been asking me, who is staying over until Sunday. If you're staying over until Sunday and you're looking for colleagues to do things with, stand up so that you can at least see each other. So anyone who is staying over until Sunday, stand up. Yep. Great. So lots of things to do in Seattle, even it's raining. I hope you have a good stay here and hope to see you soon, maybe next year.