Chinese Academy of Sciences Charlie: Cloud computing and grid exist a certain difference

It is understood that the nature of cloud computing is the pursuit of lower cost, more resources to save or resources more efficient, faster. So what is cloud computing? What are the elements of cloud computing? Where is the boundary? What is the ultimate value for an enterprise application?

Charlie, a deputy researcher at the Chinese Academy of Sciences, said the relationship between cloud computing and the grid they are essentially the same, all belong to the distribution system discipline, but there are differences, first from the system composition of the way, there is an important difference, network Computing emphasizes geographical distribution, and I now touch the cloud, is a single management operations, Geographical distribution across the management domain and single management domain implementation is not the same difficulty, the single management domain system to achieve a simple, fewer problems to deal with, this is a technical difference. In addition, the pattern is different. I agree with Miss Wu's definition, no matter what mode, all with the previous mode of application has undergone great changes, this huge change is the main source of what? For example, the ability to back-end has been greatly improved.

We are commonly used powerword, before four hundred or five hundred trillion, Kingsoft and Google cooperation, the back end has a network way to extract some words of information, the ability to improve, the reason is that the back-end capacity, storage, computing, data processing capacity has been greatly improved, this is also a pattern of differences.

Also, the pattern certainly includes the business model, you have a new market appears, everyone is working in this direction, the traditional software industry is bound to be the cloud computing model will have a significant impact, it is bound to generate new markets.

From the academic point of view, the problem of distribution system, classic issues, scheduling, resource management, consistency, security, and so on a series of issues, these are distributed systems, but in accordance with its implementation, achievable, practical simplification, so the academic and industrial will have a good combination, the original we do n years of research, Rarely on the industrial, direct use; now there is a very delicate bond.

Charlie at the same time, the cloud operating system analysis, he pointed out that we and domestic and foreign internet companies have a lot of cooperation, we also found some problems in the Internet enterprises. This is the 2009, FB speaks of its own internal business processes and basic composition, from this diagram can be found that the core of the business is such a data-processing cluster, the perimeter includes the front-end, including user information, and the back end is around this core to operate.

We are in the country, also with the domestic Internet companies have cooperation, they almost copy this structure, but we found some problems.

First, massive resource management, the scale of this cluster is very large, thousands or even tens of thousands of units, this resource management problem is what, is using such a system is too big, the system a big, the problem is exposed, once the trading volume drops, also do not know what the problem, all the defenders will go up to find the problem, This is a great challenge for technology management.

The second problem is the need for universal large data computing. This is in line with the Facebook owner's model, the domestic Internet company is similar to him, but each one is not the same, can we sum up the common to form such a platform or operating system for those who need to manage large-scale machine processing data use it?

Now it seems that the Internet industry from the business of the king, the data for the king of the transformation, how to generate value from large-scale data is a big challenge.

Another demand, is now a trend, this data I can now save, can deal with, what to do next? In order to solve the real problem, here is an example of a picture. The figure is a new architecture proposed by Facebook, Facebookpuma, and the top two are compared before and after, and the real demand is now in front of it.

Now we do the work, also around the open source organization, we did a column-by-row integration structure, which is already integrated with many open source platforms, and Facebook has already used it in production, all data has been converted or is being converted, saving 25% of the resources Other performance improvements are around 23%.

In the multi-attribute online query, we have used it in the Treasure Cube, his annual income of tens of millions of. We are multiple attribute online query, the original seven-day data capacity increased to three months, processing capacity increased by one order of magnitude, throughput increased by seven times times.

Open source software has a lingcloud, do a resource partition, resource separation.

Cloud operating system, we are faced with the scale of millions of data centers, how to manage this data center is the cloud operating system to solve the problem. As a system-wide assumption, we want to be sure that the data center across the spectrum requires extensive data dissemination and data security, of course, this is related to the business. can provide sharing and infrastructure, and can bring computing, storage and network resources dynamic aggregation, on demand, we also need to control wide area things.

Internet level data acquisition, storage, processing, Exchange to service life cycle, is off-line and online data computing load, such as search engine, data analysis, data mining, we also set up a cloud computing system evaluation system.

Another important part of the cloud computing system is the security system.

Cloud operating system Research direction suggests, Laas, flexible management of the new data center of the massive computing, storage and network resources, PAAs coverage of Internet level data interoperability.

We have done a lot of work in had doop, open source, December 3, we hired Google, Facebook some of the academic celebrities come to share, today the theme of this meeting is massive data digging treasure. We also have this information on our website.

Tsinghua University Wu Yongwei: I now do the security of cloud computing platform. I used to be safe, I don't know how to encrypt, I do not know how to do the security of the people how to see this thing, I look at the system from the point of view of the problem, the most embarrassing to me is a security issue, the meeting, a female teacher to come to use my cloud storage, she said you can see my data? I said I could actually see. From this issue, the school propaganda department to find us, said this will pull out the trouble things, the focus is that we do the system, people can see the user's data.

In fact, there are many people in the cloud to do security, there are many ways and solutions, the final implementation is a bit more dense, placed in the hard drive, regardless of user encryption, or system encryption, in addition, I guarantee the encryption of storage, you can decrypt the system, if such data, the last thing to deal with, even if not handled, in the upload download, To pass through memory. If I check in some way, the results of your core data, I can absolutely see, if I put the data inside the memory, and then slowly to push, this is completely possible, this is the core problem. How do we make sure that the data we run, especially the privacy of the morphological data, is protected, so I think the problem in cloud computing, the first is privacy, which we did a little bit of work. If the project, suggestions, I think this can be aggravated.