Facing Big Data Processing Of Multi-core Processor Cache Consistency Protocol

Nowadays, the amount of data is increasing continually. Therefore, a large number of data in microprocessors process are needed. The big data has character of low value density, which reduces the communication between the processor cores. The traditional multi core processor facing scientific computing which is not suitable for big data processing. Based on the characteristics of big data, new Cache consistency protocols should be researched.The key works in this paper are listed as follows:(1)Decrease the overhead by avoiding multiple shared copies response at the same time. In the MERSI protocol, when the system has multiple shared copies, only one copy is in shared response state, and the other copies are in “S” state. When remote processor need operates on the data, the shared response state copy for this operation. It avoids system overhead which brought by multiple shared copies when its respond to the remote operating simultaneously.(2)Decrease Cache load imbalance by copies of the state changes. In the MERSI protocol, the shared response state copy becomes the rest of state after the response operation. At the same time the requester’s state becomes the shared response state. Thus it avoiding a Cache too busy, while the other Cache are in hunger.(3)Decrease the overhead of Cache ping-pong effect by mixed write strategy is used. In the MERSI protocol when the system has two shared copies, the write strategy is write update. When the system has more than two shared copies, the write strategy is write invalidate. Using mixed write strategy can reduce the system overhead which brought by ping-pong effect.(4)The SPLASH-2 parallel program standard is chosen to compare MERSI protocol with MESI protocol. At first, the organizing methods of directory structure influence on the performance of the system and the Cache block size effect on the system performance have been tested. Then, the numbers of the processor cores and the network topology effect on the system performance have also been tested. Finally this paper have compared MERSI protocol with MESI protocol, MERSI protocol on the system performance is better than MESI protocol, the system performance improved by 3.58% on average; In the failure rate of L1 Cache, MERSI protocol was reduced by 3.18% on average than the MESI protocol So this paper basically achieved the target of design of MERSI protocol.