Yungang Bao received his B.S. degree from Nanjing University in 2003 and Ph.D. degree from Institute of Computer Science, Chinese Academy of Sciences in 2008. He was a postdoctoral fellow in the Department of Computer Science, Princeton University from 2010 to 2012. Currently, he is a research fellow of INSTITUTE of Computer Science, Chinese Academy of Sciences, and deputy Director of the Center for Advanced Computer Systems. His research interest is computer system architecture. Professor Bao has published more than 10 papers in the journal of international First-class Computer Systems Conference. He has twice won the first prize of Excellent Paper of institute of Computer Science and won the 2013 CCF-Intel Young Scholar Award. At the same time, he is the director of CCF Popularization Committee and the editorial board of China Computer Society Newsletter.
Giiso Information, founded in 2013, is a leading technology provider in the field of “artificial intelligence + information” in China, with top technologies in big data mining, intelligent semantics, knowledge mapping and other fields. At the same time, its research and development products include editing robots, writing robots and other artificial intelligence products! With its strong technical strength, the company has received angel round investment at the beginning of its establishment, and received pre-A round investment of $5 million from GSR Venture Capital in August 2015.
Cloud computing is facing a world-class problem: the contradiction between resource utilization and user experience. The question is whether it is possible to have your cake and eat it too. The answer seems to be not so simple, and even tech giants such as Google, Baidu and Alibaba are also worrying about it. We have the honor to interview Professor Bao Yungang, CCF director and researcher of Institute of Computing Science of Chinese Academy of Sciences. Professor Bao’s report on Cloud Computing and Labeling Von Neumann Architecture (LvNA) proposed in CNCC 2016 will hopefully solve this problem and become a new breakthrough in the field of computer science.
1. On the issue of improving user experience, many people in the industry attach importance to the acquisition and utilization of high-quality big data, but they rarely mention the current problems of cloud computing. Is it due to lack of awareness, deliberate avoidance, or the oligopoly of big data?
Teacher Bao: Actually, the hero behind cloud computing is data center. In China, data center is often understood as a place to store data, which is the wrong perception of data center. In the world, a Datacenter is a supercomputer. For example, Google calls a Datacenter “Datacenter as a computer.” the ultimate purpose of the Datacenter is to make it as easy for users or developers to use it as a computer. It takes a lot of technology to do that, and Google has been trying to do that for a decade, hoping to make data centers easier and cheaper to operate. So we can’t just think of the data center as a place to store big data.
In Intel’s view, the future will be the era of data centers. More specifically, there will be two models in the future. One part of the computing will be placed in the mobile terminal of the phone or the object of the Internet of Things, and the other part will be placed in the data center behind. A variety of applications will run on top of the data center, such as cloud computing, big data analytics, high performance computing such as HPC, and various network functions such as NFV. In the future, the data center will become the most basic infrastructure on which many back-end applications can be implemented.
Right now, the data center is so poorly understood that the challenges it presents are overlooked. In fact, Internet user experience is closely related to data center technology and performance. For example, when we press a button on our mobile phone or click a post on wechat, about half of our time is spent in the data center. However, there is a problem with our data centers today: there is no way to ensure user experience and resource utilization (specifically CPU utilization) at the same time. If you increase resource utilization, the user experience (response time) will decrease. Just like the traffic on the road, if many vehicles run on the road at the same time, the utilization rate of the road is improved, but people’s travel experience will be reduced.
2. Is everyone aware of the current difficulties of cloud computing?
Teacher Bao: I think there is a transmission process to realize such a problem. This problem was first encountered in Google. When the development of cloud computing or data center reaches a certain scale, this problem will be manifested. This is a requirement delivery process, and Google identified this problem early on and has been addressing it, calling for architectural innovation and hardware support. Domestic leading enterprises such as Baidu, Ali and Huawei have begun to invest a lot of technology to try to solve this problem. In a few years, more enterprises will pay attention to this problem. At present, this problem is relatively cutting-edge.
Giiso information, founded in 2013, is the first domestic high-tech enterprise focusing on the research and development of intelligent information processing technology and the development and operation of core software for writing robots. At the beginning of its establishment, the company received angel round investment, and in August 2015, GSR Venture Capital received $5 million pre-A round of investment.
3. How will labeling Von Neumann architecture (LvNA) improve THE CPU utilization of tablets, phones and mobile devices?
(Note: During the interview, Bao adjusted the question to: How does LvNA improve the CPU utilization of data center servers?)
Teacher Bao: Actually, this question is not appropriate, because ordinary users often do not pay attention to the mobile phone in their hands, how high the CPU utilization rate of notebook computers. However, big companies like Microsoft and Google pay a lot of attention to CPU utilization. The problem of low CPU utilization is more from a data center perspective.
There is a paradox: on the one hand, the cost of building a data center is very high. For example, Alibaba’s data center in Zhangbei cost 18 billion dollars. Microsoft now spends 30 billion dollars on global data centers. Google spent 11 billion dollars on building data center infrastructure in 2014 alone. Data center, on the other hand, the CPU utilization rate is extremely low, amazon has millions of servers, for example, investment of tens of billions of dollars, but, a few years ago, a report reveals its business in cloud computing data centers to run on the CPU utilization of teens, the utilization rate of the equivalent of what is wasted, but it can guarantee a better user experience.
Over the past few years, these big companies have put a lot of effort into dealing with this problem. In order to solve this problem between resource utilization and user experience, Google optimized the entire software stack of data center, from the bottom virtualization, operating system to the upper distributed architecture. But in 2015, Dick Sites, a fellow of the American Academy of Engineering and Google’s data center expert, acknowledged in several reports that it was difficult to achieve performance isolation between applications on a shared server, and that Google had done its best to optimize its software and needed hardware support.
We propose labeling Von Neumann Architecture (LvNA), hoping to provide the key technology to solve the above problems from the hardware level. Using traffic as an analogy, LvNA aims to ensure that critical vehicles (such as some very demanding applications) can pass through already congested roads. By using the tagging mechanism and enhancing this control mechanism, some applications requiring high experience performance can be guaranteed to run at high speed even when CPU utilization is improved.
In fact, we also have labeling mechanisms in our lives. On the road, we will see ambulances, fire engines, police cars, private cars and so on are painted differently, which is a kind of labeling mechanism. And then, we have traffic rules that make sure that when those cars — ambulances, fire engines, police cars — honk, they move out of the way, and if everybody follows those rules, they can get through in an emergency.
We wanted to apply traffic rules to the computer architecture, to label key applications, and to create rules of the road to ensure that the most demanding applications get through first. In this case, some key applications are not affected while improving CPU utilization. In this way, shared hardware resources are allocated on demand and performance is isolated.
4. Computer is composed of hardware and software. Does the labeling Von Neumann architecture (LvNA) proposed by you have higher requirements for software?
Bao: Actually, every kind of hardware needs corresponding software to adapt, and software needs to give play to the functions of these hardware. Throughout the history of computer systems, the hardware provides new functionality and the software subsequently changes accordingly. After the computer architecture with the labels, can be implemented through the label isolation performance, enhance the control, after this, the software also need to undertake corresponding adjustment, which requires the software can be associated with the set label, such as the label for the kind, how to develop different rules according to different labels, this is what we usually speak software definition.
5. What are the advantages of labeling Von Neumann architecture (LvNA) over existing computer architectures?
Bao: The core advantage of this new architecture is that it can enhance the control ability of computer software to the architecture. Originally, some applications in the use of resources inside the computer present a disordered sharing state, just like the vehicles on the road do not obey the traffic rules, like a rush through, which will reduce the utilization of resources. Traditional architectures do not pay much attention to this problem, but in fact, it is a very core problem in the data center. If this problem is not solved, the resource utilization efficiency of the entire data center will be greatly reduced, the cost will become high, and the user experience will not be improved. We propose a new architecture from such problems as the departure, in view of the data center facing such a severe problem, after analysis, research, find the crux of the problem – disorderly sharing, then slowly to solve this problem, in the end, after a summary to put forward the new architecture, labeled, von neumann architecture according to the man.
Both academia and industry are paying much attention to this new architecture, and Professor Bao is confident that the label-based Von Neumann architecture will revolutionize the research and application of computer servers in the future.