The flaws of cloud computing

This thing gives me a headache. I know that cloud computing is very important and is being pushed into the plans of various corporate governments. But I always think cloud this dongdong and the familiar system compare. It seems to be a short cut. Of course, perhaps because it is still in the early stages of development. There are a few words to sum up about its flaws:

1. It has made full use of hardware resources, and it has also consumed a considerable portion of its hardware resources. When I see a certain computer room each site occupies a server, think cloud computing is very very necessary things. In fact, those servers are very electricity and the server in the Web site from I began to use the Internet to today, never logged in. Those websites are well-known .... it. When I use my own machine under Linux full virtualization out of an XP. I immediately feel that virtualization is a lump of ... No, it's a lump. Machines are not permanently overloaded, and full load machines are not virtualized, and inexpensive servers are expensive for me.

2.GFS is really bypassing virtualization technology. Hash, sorted plus two points. Very good, very strong. Just bigtable + mapReduce + GFs is sometimes less flexible than MySQL. How to say MySQL, SQLite also support a transaction ... You want to create a transaction that requires you to add a user lock code. How do I want to add row-level or column-level locks ... Can consider to modify the source code ... Besides installing Hadoop is as flexible as installing MySQL?

3. Lock or not lock, this is the problem of distributed headache. I have a headache, too. Lock bar, low efficiency, no lock, check consistency. So compromise, only write lock. The conflict between writing and writing is solved. The conflict of reading and writing continues to remain. So keep writing the momentum, not the intention to read successively. Anyway, the program is not wearies, read more than a few times it doesn't matter.

4. Is it really worth the rent? Is it really a savings to rent a machine? I think it only applies to short-term projects and long-term projects. Facebook is renting part of Amazon now, but if Amazon's price changes, it's going to go up every year. Do you understand. ...

5. How to stir a lump is still a lump ... Or a lump of ... MIPS (million instructions per second) Small machine, I am difficult to virtual machine MIPS large. (Do you want to have each server automatically overclocking?) ... This seems to be a good way to destroy the cloud ...) So the individual thinks: the cheap machine cloud does need to find the ideal large-scale parallel distributed computing model MapReduce like Google. Otherwise, the performance will only be reduced if the virtualization method is used. MapReduce model is a commonly used hash+summary in distributed system, which is very effective for the poor performance of special optimization calculations.

About the future:

I think the cloud is still very promising. Or what am I going to do with it? But clouds should develop into heterogeneous networks, and virtualization technology will have to be developed. In addition, the programming language has to be improved. In addition to say so much ...