Panel discussion: Emerging computing technologies in academia and industry

foreign next in this track we've got a panel emerging Computing Technologies in Academia industry what is the next big thing in cloud computing what are the disruptive technologies that are just around the corner how can industry and Academia work closer and better together to answer these big questions we brought together some big Minds from Microsoft and Academia these are the people who are working at the frontiers of the cloud cloud intelligence and its applications our distinguished panel includes Yuan Yuan Zhao a professor at UCSD Dave moltz from the Azure core team who runs as your networking John sheenan from Azure Edge and platform and Jim clewine from Office 365 and our host for this next session will be Thomas Cara Guinness a principal researcher from the cloud systems future team here in Cambridge so let's go and hear what they've got in store for us welcome to the fifth session of the future of the cloud track which is a panel discussion my name is Thomas karayani and I'm a principal researcher at the cloud system Futures group at Microsoft research Cambridge UK I would be moderating this panel sessions which for the next 45 minutes aims at discussing emerging cloud computing Technologies with disruptive potassium and how to align efforts across academian industry I would like to believe that we have the most excellent panel composition for this discussion so let me quickly introduce the panelists y1 zoo is a professor in Mobile Computing at University of California San Diego since 2009. here area of expertise includes Cloud reliability data center management and Cloud security she obtained share Masters and PhD from Princeton University and she is an ACM fellow and I triple fellow a Sloan resource fellow and the winner of the prestigious ACM Mark wizard award in 2015. young one actually in parallel to share academic careers he has also co-founded three companies one of which was acquired by VMware so besides being a leading professional young one has does a lot of experience interacting with a broader industry John sihan is a corporate vice president distinguished engineer in the cloud and AI business at Microsoft he runs the health and standards team which owns the health infrastructure in standard Cloud architecture systems for Microsoft previously he was responsible for the architecture of the Azure iot platform and services outside Microsoft zone is also very active in various ways in encouraging women to pursue careers in stem and volunteers time with Ignite worldwide an organization dedicated to achieving gender equity in stem Dave mulch is the engineering leader for the Azure networking team responsible for developing deploying and operating the software and network devices that connect Microsoft's largest larger services including the other public cloud and Microsoft 365.

Dave is essentially responsible for the team that designs the cloud scale networks and data centers that provide petabits of connectivity with higher liability all the way from the optical systems and up Dave has also been a member of Microsoft research in the past with many Publications Awards conference Keynotes so he's very familiar also with academic research and last but definitely not least Jim clewine is the Microsoft technical fellow and currently leads the team responsible for the cloud services supporting Microsoft teams other communication services and Skype he has worked on Office 365 cloud services since the very beginning and was actually responsible for many of the initial designs architecture and Technical approaches that moved office servers essentially to one of the world's Premier hyperscale cloud services thank you all for joining the the panel again and for participating now before going into our main discussion actually I would like to have a quick warm-up round and I'm going to ask uh from you just two words so these words would be what are the two words that come to your mind one word for describing the biggest challenge of a future cloud and one word would be for the biggest opportunity for the future cloud and we can go in the order of a presentation that I just used maybe you can go from Yuan Yuan to John to Dave and Jim so young one over to you two words so the first word in terms of challenge um is it a complexity um so do I need to explain is or you I just give you two words just give me two words yes we can we can go over them so the challenge is the complexity and then the um the second wording opportunity is uh really is the data okay John and efficiency darn John you stole mine but I'm gonna double down my two words are scale and scale uh Johan Johan took one of mine I think the biggest challenge is complexity uh and the biggest opportunity is enabling him wow okay so there does seem to be a lot of both agreement and a lot of diversity in the uh words used so let's let's keep this word in the back of our mind and let's let's move on to the main main discussion uh so I would like to start in in reverse order now and go through the cloud stack from the bottom up bottom up starting with the infrastructure so perhaps uh this question is mostly for a day even team but but uh John and Johanna one feel free to jump in so we have heard talks in the earliest session of the future of the clouds track with research exploring very different approaches as to how we do compute networking and storage for example uh we have we have heard talks exploring Optical Computing or storing data in glass with project silica or innovating across the whole networking Cloud stack and in my mind these types of Innovations and these types of research is trying to address the fact that traditional Technologies such as CMOS have stopped scaling and I believe this is well accepted now now my question to you is is this the right time for this types of Technologies to feature in the data center and if so what does this path look like how do we take these disruptive Technologies and make them a reality or or is it still too soon for them so maybe we can start with uh with Jim so thank you Thomas uh I I think uh that it is a well-established fact at this point that Denard scaling and Moore's Law uh if not all the way dead are certainly getting closer and closer to it every day dinard scaling is is long dust and uh and Moore's Law is more debatable the emerging Technologies I think are things that we have to start to explore and we have to start to incorporate into in the data centers if we're going to continue to do things like scale one of the words that we heard for both challenges and opportunities from from both of us and if we're going to try to allow uh us to make the solutions more simple we have to have changes in the underlying technology platforms the underlying foundations of computing and foundations of data centers to enable the really complex parts of the data centers which is the software to continue to be as simple as we can make it so the right time to innovate in these new technologies is always we've always got to be on The Cutting Edge of doing this if we're going to continue to scale if we're going to continue to offer the ability to be simple you know while uh while Moore's Law is dying and Denard scaling is dust that doesn't mean even at the lowest level at the Silicon level that there's not room for innovation uh you know I'm I'm cautiously hopeful a lot of the work that's going on uh on the science side for graphene will play out we'll have a whole new class of computing devices coming out I'm very excited by improvements in Optical and and and the work like silica and other Optical work like it uh because no matter how much uh we would like to we still haven't kicked Einstein's but that speed of light is still as close as we can get uh to fast and anything that gets us closer to the speed of light is a is a good thing I love the Innovation coming out of AMD on shiplet technology so even if Moore's Law is dust it doesn't stop us from continuing to scale the number of cores on processors and and to do it in an economical way uh because of their chiplet architecture so I think there there are tons of ways in which there's still Innovation at those lower levels of the stack and and even moving one click stop higher from the Silicon level into the computer architecture level uh if we look at it a lot of the fundamentals of our computer architectures uh have remained stalled for you know 20 years maybe I mean a rack of servers is still largely well the same architecture that a server was 20 years ago where you've got the the processors and the buses and the dram all in a package solution tightly bound together we haven't really revised the basic computer architecture to reflect the fact that we're not dealing with pizza boxes stuck in a in Iraq we need to realize we're building and operating hyperscale services and look at Innovation at the computer architecture level to to Really Drive some of that forward question some of the basic presumptions that have kept the computer architecture really stale for 20 years uh you know what one of the things that I love about all of the work that I do with uh with research and with Academia around the world uh is that research in Academia explore the realm of what's possible less hindered by what currently exists they have a tendency to think outside of the box where Within straight line development and straight line engineering we are relentlessly incremental and and sometimes you have to stop being incremental and you have to allow the giant leaps that you can only get by crazy outside the box Innovative what happens if we change a fundamental presumption thinking that we tend to get more from research in Academia so I'm very very bullish on the future for the bottom lines of the stack and the thing that would make me even more bullish is if we had more and more researchers who were looking at those bottom layers of the stack at the you know stodgy old uh boring systems level of the stack and saying how can we really innovate there you know anytime something gets to be really static and anytime something becomes accepted as the way things are done that's the time we most need the crazy thinking and the research and the innovation uh a great pass to maybe maybe ask David you know just work out we're going to work and kind of innovate and we can think outside the box the question is how do we bring this into a you know as people mentioned like a complex system which is of a huge case so what would be the best way to do that what is the password what's in this okay great so you know Jim I completely agree with a lot of the things that you were saying there maybe just put a little bit of a networking twist onto that um I said the challenges scale and the opportunity is scale because the cloud and the future of computing is really about Limitless Limitless computation right how can we take all these resources and make them look to a user as if they're one fungible infinitely dividedly infinitely expandable set of resources that they can apply to whatever project or problem they're trying to solve um you know to start with the first premise though never bet against the device physicists I mean those folks have come up with Innovation after Innovation so yeah all of us should keep our computer architecture hat on and look to see what's happening at the various lowest levels of innovation some really radical things are coming there in multiple fields that could then change what we as computer scientists have to do pretty dramatically in terms of solving this problem um yeah let's take it for granted that you know the simple scaling that we've been getting from Moore's lives over what does that mean it's going to happen well great we are going to replicate more units of computation and that's how we're going to get more scale which essentially means all those problems of parallel Computing that we thought about a lot several decades ago and then have been able to largely put down and ignore for some period of time are coming back at us um so I actually start thinking about this problem essentially at the programming model level uh the kinds of programming models that we've been using are not very amenable to the type of infrastructure that they are likely to be running on in the future and that's really worth thinking about storage in a different way thinking about serverless or stateless Computing things like Azure functions those are very powerful programming models that make it relatively easy for humans with our demonstrated limited ability to think about power parallelism to actually write code and programs that are going to work as we then start to take that you know what we want to expose to the programmers and the uses of our systems down to the next level um like Jim was saying computer architecture becomes really important but I want to call it a couple of things in particular the coherence architectures like what is that memory hierarchy and to what extent can we simplify the programming model for the language designers as well as for the programmers by maintaining coherence at the architectural level that's really hard because as the scale goes off and the number of things we're trying to maintain coherence over increases it's easy to lose a lot of performance this is fundamentally a systems and networking problem why do I say networking well okay I'm a networking person but imagine what happens right so essentially we're going to get more scale in our computation by adding more endpoints into our data centers or into that sort of computing infrastructure running on top of all right so interconnect becomes super important then right that interconnect is the network um I think we're going to have lots of different kinds of interconnect uh yeah the speed of light is is the ultimate uh uh speed limit out there um but there's also factors of like how much power it takes to drive information from one chip package to another from one box of Chip packages to another box of Chip packages typically the further you go uh the longer it takes but also the more energy it takes and power consumption and the ability to remove heat uh from data centers from the components of the data centers is really becoming a limiting function a limiting factor on how big we can build these things um and so for those reasons you know we're probably going to end up in a world where your data center that set of fungible resource is going to look like a very non-uniform architecture and again the programming languages folks you're going to have to figure out how to expose that to users and we as the systems in computer architecture gonna have to find ways to try and simplify as much as possible um what that also means is the networking space lots of room for Innovation because we are going to be managing different kinds of interconnect but we want those different interconnects to work together seamlessly that may be data buses specific for memory different ways of managing storage both local and remote as well as then the ability to get data between lots and lots of specialized code processors gpus tpus dpus ipus you know star pu is the name of the game and systems and each of those things has different requirements for how they need information to transfer between them and and what assumptions they want to make about the interconnects underneath them so there's this really interesting interplay between the people designing those co-processors and essentially the transport libraries and the interconnect networks that we're going to build to move data around inside the clouds of the future and then I I guess the final point I want to make on this is you know we can't forget that this is a physically embodied thing and so work in Optics super important again can reduce the energy per bit it takes to transfer data but also gives its way to then physically cable these sort of monsters that we're creating um these ultimately have to be reduced into three-dimensional spaces where cables can be run in ways that we can actually build a facility that can be scaled out that can be maintained that we can grown over time people often like to think about oh the supercomputers of the days of yours those were typically built all at one time as a special bur uh special purpose computer the world of the data center in the cloud is continual deployment and what that means is I always wanted to be deploying the most efficient newest technology that means if I look in our data centers today I still have some stuff running at 10 gigabits per second servers at 10 gig networks that haven't been decommissioned yet I've got stuff at 40 50 gigs 100 gigs I've got stuff at 200 gigs and 400 gigs coming and so there's going to be this heterogeneous mix of the physical infrastructure but again our goal is to enable true multi-tenancy where we can take many different workloads running on top of this heterogeneous infrastructure while still giving every user The View that they have a homogeneous Cloud that is perfectly isolated to them so that there's you know what one user is doing doesn't impact anything that other users doing and that creates a whole bunch of interesting systems challenges as well from everything we saw with some of the cash leakage behaviors meltdown inspector to simple performance isolation you know anytime that there is a place where two different workloads come together this potential bottleneck where the behavior of one Cloud user could impact the performance seen by another we have to solve all those problems as well uh actually they are going to pick in uh one line that you use that you know humanly have this limited ability in understanding these sort of complex systems and I'm going to to kind of try to move up the stack and ask uh you and why uh what do you believe is the role of AI operations in sort of these complex Cloud scale systems you know are we going to be relying on AI completely to run these systems or are we going to have humans in the loop and how does this interaction look like you know what are the challenges for AI to ensure like that we have a truly ubiquitous always on cloud actually um like as I mentioned there's all this heterogeneous and also constantly evolving right with all the new devices it's really beyond the human's capability to kind of manage this actually um I find that previously I founded this problem but you know when I did the previous startup it was really the cloud the back end the management and infrastructure there is always fairly complex where the all the same new emerging technology and sometimes they really use this completely different kind of a set of standard or API so manage them actually especially like do troubleshooting or like a kind of our security for insects is actually really kind of challenge um I think there's a whole lot of effort in this on the infrastructure side so the current startup actually is one I'm doing it on all Cloud applications so I realized that the things is much much worse because of the human actually like it these stays a lot since I'm also from Academia you know like application like developers the Cs computer science students they're moving up right so right now I'm teaching OS classes um in at the UCSD what is no longer required for students so in this case what does it mean right so then that means like we're going to produce a lot of developers they don't understand the back end like all this basic low-level stuff so now you imagine they're going to join some company do e-commerce even though they're not like a part of like a big you know like a Microsoft or Amazon those are like a clouded service provider but they have to manage the cloud applications and which also use a lot of like a database server um like kind of also like open up you know Object Store many kind of different boxes that they also use a third or seven a third-party service providers right so it's all like a kind of intertwined together and uh I found actually even sometimes my master that was you know like a developer with a master's degree they really don't know how these boxes operate with each other and that once something goes down it takes you know quite a bit a long time to find one of the third party Library I had some issues because you know because you need actually to communicate across the cloud to the service provider right some kind of application Level service provider for example the one provide you a map right um so I'm actually quite a really worried so but I think the trendy is going to go this way because people students want to learn like a data science AI so something more high level right so the the Challenger I think it's then the like management and naturally gonna go more towards the Automation and the AI is a way of automated things so that's actually the kind of way and maybe kind of uh for the human part I think it's really the another product that I have been very passionate is that there's a human interface because uh like our mobile phone we worry about a UI ux but in the data center management or some kind of management thing I think UI human interface interaction is not a bigger kind of like a kind of requirement so as a result it's actually once many times the way we to manage the things is to read a lot of unlocks right kind of what's going on with to be able to understand so I think it's like on one hand we should we should use automate as much as possible using any kind of intelligent like AI algorithm or some Automation and on the other hand for things with Mother fully automated we really really need to emphasize the kind of human interface is any way we can visualize in the way for them to see be able to easily see where is the problem and so I I think data is a kind of the part I think I'm actually particularly worried about another you know like the infrastructure level I'm more way better the application Level right now I know realize you know how I'm prepared that those application developers are I see absolutely can I jump in there Thomas with just one comment because uh I love what you said there about the the people uh more people learning systems and learning operating systems because I I think the the real differentiated skill set for students of the future is going to be people who understand and can reason across the entire stack because of the complexity we started talking about and because of a lot of the things you brought up so you know keep keep training as many of those OS students as you can so don't seems I think you are kind of the best position to talk about health and infrastructure so what what are your thoughts given the picture that uh why I painted on how Sooners sort of move up and sort of ignore everything below yeah so I mean as a long time operating systems guy breaks my heart um operating systems are really fun to learn about um and it's it's kind of unfortunate that it's uh it's not even a required course anymore um and I do think that is going to cause major challenges um I already see it now um you know with uh sort of new students coming into um you know into Microsoft trying to debug very complex issues and and sort of hitting a wall where they just don't understand how that level works and those black boxes can be really dangerous at scale um but as far as operating our cloud like I mean so I own the health and standards team um and part of my team is building the AI Ops platform uh for us for operating our our Fleet and you know when I look at the scale of telemetry we get right now um just from our clouds we get over I guess over now we're at over 500 petabytes a day um there is no way that humans can look at all that data and and make any sense of it at all right we're way past using humans to operate at the scale of our cloud and so you know my team is constantly looking at what are the ways that we can take that data um and you know reason about it and figure out automatically things like something as simple as service Health right if you think about just take one service that's off that's running um and just try to figure out is it currently healthy um binary thing right I mean we've got millions of customers running on that one service and some of them may be having a great day and a couple may be having a bad day and being able to go through the data and figure out like what's going on um you know another big one is introducing change to our Fleet right so we you know we I don't know how many thousands of changes we push hey any one of them can cause customer pain and customer impact figuring that out detecting it before it happens you know when you're trying to sift through you know over 500 petabytes of data like it's it's just it's the the numbers are are just mind-boggling and we know that we're growing like we're constantly having to plan for 10x growth right so the the challenges that we have today are a small uh piece of the challenges that we're going to have in the future um especially you start thinking about how we're growing you know as Dave was talking about you know we're we we take we take a unit of compute I think of a unit of compute as like a data center um and we just keep stamping out more and more and more from them and the rate at which we're stamping them out is going you know up and up and up um and uh and so that means you know when we talk about complexity you know we talk about operational complexity and operational complexity is actually a a product of the the you know the overall um workload that you're running times the number of places that you're running it um because if you actually look at these systems you know this they sort of look at each one independently because they have to for fault isolation reasons and so you know our our complexity our operational complexity is just going up and up and really you know Automation and AI is one of those technologies that um you know currently there's tons of promise with AI but there's some stuff it's really good at and that happens to be the things that we need it to be good at um so you know in many ways in in kind of the health space we're fortunate that um you know that we're we're you know that that there's so much invest going into AI right now and then when I think about like the you know the future of these data centers um yes we're trying to scale and scale and scale but I'm also starting to see a lot more specialization right in you know inside of our Fleet we're starting to use you know all kinds of different cards that are specialized to handle a certain piece of the workload to free up the general compute for the customers um start to see things like tpus you know where we can have specialized compute that says yes Moore's Law is not scaling the way it used to but that's okay because General compute isn't the only way to go there's also specialized compute we're looking at all of that as how are we going to deal with these you know this ever increasing amount of telemetry and data that we're getting like we're going to have to go specialize to to deal with it I see I see thank you for uh for for all these insights uh I think I would like to spend the last 15 or so minutes on uh of the session discussing on how industry and Academia help each other on this sort of very difficult journey through complex scale as you mentioned so uh if you enjoy I think you're coming from Academia but you also have a sort of experience interacting with industry uh I guess academic research is not as dim mentioned not bound by the restrictions and constraints of existing systems and existing clouds and how they run and so on and they can offer very different perspective to new technologies so in your experience what is the best way to get some of these insights towards the industry and not just keep them in accordance in Academia where we like them in conferences and so on but they never made it to reality so what what are your thoughts on this one so I think one thing is like a active actually the good thing is we uh typically faculty have a sabbatical and there are also students they can do internship uh like for example I have uh probably three PhD students who did extended internship at Microsoft action without you not with Microsoft research so I think like one student stayed there as a sort of intern for one year I thought actually sometimes like a three month of Summer internship is not that effective because ticket and probably two or three months to even know understand the system and how things work where are the data um so I think sometimes in this kind of way if there's some kind of way to do extended internship I think especially for PhD students it's actually they're they're they love this opportunity also for the professor they love this kind of collaboration because uh due to the constraints because if the we just purely collaborated this way um like the students don't cannot access the data right they're not like the employee of the company so for like a million time in uh in IP issues it's better for the student to be hired as an intern so the students can access the data but the other Professor usually just advise the project collaborate with the Microsoft you know people from Microsoft together I found that it could be very effective because uh this both of the PhD students and the faculty that they can do things a little bit extreme in terms of like a phone number David talk about the programming languages right so we actually come completely propose a language or another word that much but adoption like uh but we will be able to explore different kinds of languages and then that maybe not be needed to use the kind of right away but it could be some in the future a near future some elements of the languages could it be putting into like the kind of the business actually the the real world right so that's all kind of one thing of another part of a thing is like a professor actually can't do sabbatical for some time or one year or even extended the time um I found that for me because the benefit of dual startup um actually really beneficial for my research because if I don't go to the industry I would never never realize the many many of the problems so far something especially maybe since we work on computer systems I think we 100 State University we're not going to know what a problem out there so I found out I gave like spending I have a like a researchers from Academia to go to Industry spend some time they can take the problem back and then they'd be able to this way when they come up with some solutions here is irrelevant you're not completed off the chart in our own toy mind to you know think about the toy problems right so I think that is this way is that those kind of solutions are more easily applied back to uh industry like the the like the second startup was actually what you mentioned a quite about VMware was a video and we discovered I was like in the toilet to do debugging like a 100 bucks and source code but didn't actually want to talk with industry they say it's um you know many times actually they said we have tons of buckle I don't want you to tell me we have more bucks so can you help us at troubleshooting diagnosis and many times we cannot reproduce the problem we can only look at the logs right that's how we people diagnose and analyze and then you say any way you can help us in college and analyze the logs so that's actually we took I took the problem of course we use something like AI all kinds of techniques some is actually so we will come back to be built like a plug the product is some of the are not algorithm not accurate enough we have to remove them because once you apply to the industry accurate is super super important you cannot introduce so many false positive in troubleshooting so I think that's but then part of our work to solution works because it came from the industry and the problem it started from the industry so like a part of the things you know we explority cannot apply to a product right away but you know a big part I can actually become a product so it's still right now used in data center to do this kind of like data center monitoring like health and monitoring security forensics yeah so I feel like a Visa kind of ways a collaboration like you know definitely other people can jump in I mean um I'll say like I've noticed in the last five years that this relationship between like at least Microsoft and and universities is starting to change um in you know in the space that I play in um we now directly talk to the professors and this and the students about the challenges that we're having and saying hey do you have anything that can help and in some cases they already have something and then in other cases they're like oh interesting problem um often they're like can you give me data in a lot of cases we can because it's our data um you know it's about our servers it's about our Hardware um and so yeah so that relationship's really changing like you know my team sponsors uh um a partnership um with Stanford um that we participate in their system X Program um and that's one way that we get access to to the professors and the students um to sort of start to to collaborate at the level of not hey someday it would be great but like here's our current problem um you know and here's what's happening and it's going to come in the next few years you know what can we leverage to solve it because we can't wait you know we can't wait 10 years for you know something to happen like we need it now right Dave I guess you also have experience in both camps research and and Industry and I guess more kind of product side roles so what are your thoughts on that one well yeah I've personally gone back and forth between research and production land a couple of times um like you on one said I the the human factor is very important so doing internships doing sabbaticals in industry is probably still the best way to really get deeply embedded in some of the challenges that are going on and during those types of um sort of visits you know long-term visits to to the production companies we're typically able to include people and show them all the data um well John said some data we can share because it's essentially about our systems um job number one for us in the cloud is safeguarding the customer data that's been entrusted to us we work very hard to make sure our employees and staff don't get anything that would give insights as to what our customers are doing on top of our Cloud um and you know we're making big Investments of confidential Computing to sort of strengthen those guarantees even more and more so I have had you know I've been part of conversations where people have asked me hey can you give us this data I have to say no even networking data which you might think is like oh it's just traffic utilization but detailed information about traffic utilization can say quite a bit about what a customer is actually trying to do with that cloud infrastructure um so that's a challenge that we're just going to have to operate from as has been mentioned sort of consortiums or um uh collaborations between industry and specific research groups I think those have been very fruitful one of the things I'm missing though is sort of uh places where people from multiple companies and multiple research groups can come together uh to work together um you know Microsoft does do a lot of work in the open source space we started Sonic software for open networking in the cloud which is an open source operating system for routers precisely because it eliminates a lot of the intellectual property issues and concerns um be nice to see if we can do some things like that In This Cloud infrastructure space as well um you know although we've compete fiercely Microsoft does with other uh Cloud hosting companies there are places where it seems like we Face similar challenges and collaboration even among the cloud providers as possible if there was this kind of neutral ground that we could all come to participate in and then and then collectively draw insights from um so those are some things I'd love to see happen one final comment um is that I you know there have been a lot of statements in fact oh it's impossible to do academic research on the cloud unless you have something to scale of the cloud the cloud is all about scale um and there are certainly problems where that is true however some of the things we talked about today like programming languages like architecture like even the details of how we get high quality of service and tune buffers and specific routers and switches those are things that do not require complete access to your own cloud in order to run those experiments so there's a lot that can be done not having a cloud and then there's a lot that can be done running on top of the clouds that are already out there and so you know I I think all the cloud providers have programs by which we can make VM hours and other compute resources available to academic researchers really encourage everyone to take advantage of those because that essentially gives you access to largely the same infrastructure that we're building on top of as well um and so part of our job as Cloud providers is to make the full power of these platforms available to everybody and that includes academic researchers I mean I I think uh Dave gave us gave us a great setup there with his final comment to bring us back to where we began which is what's a lot of the opportunity uh in the cloud and if you remember my word for opportunity was enabling and I picked enabling because the cloud really allows us to democratize access to the massive amounts of compute and the massive amounts of data that that's necessary to do research to do Cutting Edge Innovation and uh and really uh creating consortiums creating projects like YY and John were talking about and then using the cloud to enable uh them to work with the data to work with the scale is a great way to do this it doesn't have to be Consortium so it can be things like what my current team does which is challenges we have uh we have challenges that we issue where we provide data and objective metrics and say who can come up with the best new AI powered video codecs and we're looking to to things like our team's product and saying hey we could really use that to reduce things like Network bandwidth and to allow networks to scale as Dave was talking about and we have a challenge and say hey create a group and compete in the challenge see if you can beat the objective metrics on the data doing the same thing with with noise cancellation and and eliminating background noise there's so many different ways that we can collaborate uh and and uh I in my own experience everything YY said are the best ways to do it which is form relationships because in the end clouds are about software and services and software and services are about people they're about the relationships between the people and having great relationships between Academia between research between product and Engineering groups is the way we're going to drive Innovation here okay thank you I believe we're getting close to kind of the 45 minutes so thank you all for very interesting discussion uh any last words that you would like to add from anybody on the discussions I'll just say I mean the cloud remains such an interesting and challenging space I'm sure there's a pendulum swing uh where pendulum is qualifying towards cloud and making that work you know 10 years from now it'll probably be swinging back towards dust Computing or something like that uh but but this you know I think there's so much exciting stuff happen uh as the world continues to to innovate and Unleash the Power of clouds and the size of the problems that can be solved using these massive pooled resources yeah and I'll definitely make a plug for operating systems um I think Jim uh mentioned earlier the places where you sort of feel like oh that's done we're good there let's move on um those are often the places where there's the most opportunity um and you know I can tell you uh you know my team also tries to keep the you know like the one and a half billion Windows PCS around the world healthy um uh we're nowhere near done with Innovation and operating systems um so I really hope that that you know more research continues there um there's and and more more of the students that come out of universities come out with that skill set because it's vitally important to understand that if you're going to try to troubleshoot anything foreign total plus one to to both of those especially John's comment that uh that were far from done here uh I I would like to think that we've barely begun not that we're done in any of this yeah so follow like all you guys comments I think I also hoping you know like a industry experts are like you can come to Academia to give a talk and share the challenge is the exciting things happening in especially on the back end of the cloud the infrastructure because I really wanted like you know you guys helped me excited the students besides the fancy you know upper level things there's a really really cool things on the cloud infrastructure side on the cloud side so I think I think this way really hope like we can produce we can attract more students to have those kind of skill sets willing to you know like learn those kind of skills as a you know at a university send a paper to you know prep them for the workforce sign me up cool I speak for all of my peers not just those on this call when I say yeah we would love to do that uh anytime you want us to come and give a talk to any of your classes uh give give us a call your students are the Future Leaders of of the cloud and Industry tomorrow and and we want the best leaders we can get thank you very much for this insightful uh conversation and I hope the audience found it as insightful as I did thank you very much all for participating in this panel thanks take care what an amazing panel session and thank you to all the participants in that panel what an amazing set of topics discussed and the Deep Insight that we've just heard from them is just it's just fantastic and with that it actually brings us sort of to the end of this uh Summit track I want to first of all thank all of you for spending the time listening and being with us and and participating in us in a in this track with inside the research Summit I want to thank each of our speakers they've put a lot of effort in and they've thought a lot about what they're talking about and they've made it extremely interesting and exciting thank you I think as we went through the track we've heard about aim a new analog Optical computer a new way of solving these hard optimization problems which I'm sure you'll all agree with me sounds extremely interesting we heard about space and about pushing from our planet out and taking Azure from being on this planet to being in space as well thinking what we can do with satellites how we can build services for these satellites it's been fascinating we've heard about project silica in an update project that we've seen go from that ferment phase where we're trying to understand what a technology could be into that takeoff where it starts to make sense and it's really exciting to see how it's coming together as a real storage system we then heard about AI Ops and that sort of challenge of just how do we make the cloud more efficient more reliable more controllable easier to operate more manageable and how AI is going to change the way in which we do it and then of course there was that panel which again thank you to all the participants it was great now for all those researchers doing and thinking about the cloud I just sense that now is an opportunity in fact more than opportunity we have a responsibility to think about the future so much that technology and so much of what we do is beginning to get to the end of its life we need to think what the future looks like and how we are going to enable it it's critically important over the last 20 or 30 years we've seen so much progress and yet looking forward it is harder to see how that's going to happen without us working more together across many different levels between hardware and software co-design thinking of the services how we make the services more efficient how we get more performance from the hardware that we have and how we create the hardware for the future this is a time of immense opportunity and responsibility and I'm looking forward to seeing what you all do as well as what we do in the coming years so with that I just want to say thank you again for attending and I look forward to seeing you again next year that's a wrap for day two of Microsoft research Summit I hope you've enjoyed another day of presentations and discussions among researchers at Microsoft and our colleagues around the world we'll reconvene tomorrow for our final plenary session on how research is empowering medical professionals and researchers in the health and Life Sciences followed by our final day of tracks and sessions see you then [Music] foreign

2022-10-29 04:17

Other news