| In current computer system, there are eminent differences in speed between the several main components, typically including CPU Cache, RAM and Hard Disk. Their speed order is CPU Cache> RAM> Hard Disk. Their speed discrepancy is as exponential decreasing and their capacity is as exponential increasing. Thus, in order to get the best performance, operating systems usually add a Page Cache mechanism to speed up accessing the files.With the explosion of the application of big data, traditional cache system could no longer meet the critical demanding for the performance in reading and writing big data. But the fact is the reading and writing performance has been a vital factor influencing the whole performance of the network, and is being one of the hottest research fields. The I/O-dependent applications, namely database, always wait for new technologies to break up the bottleneck. Before, the huge high-end storage devices weight several tons. Moreover, they are always much more expensive than the hard disk. Its additional value is at workload balance, multiple-path error tolerance, firmware optimization and data protection.With the appearance of SSD, the above problems have been solved better. SSD converts the machinery disk to electric product, which owns higher performance, less power consumption and faster response. In this case, Facebook implemented SSD-based Flashcache/Bcache to take the challenges brought by the big data However, this cache could only serve local file systems, and in distributed scenarios it brings no improvement. Although there has already been a kind of SSD-based cache system which could work in distributed environment, it has two limitations:1) it only serves read-opened files.2) it is only equipped with write-through cache, but no write-back. These limitations seriously confined the write-intensive clients deployed on NFS.This paper proposed a kind of write-back cache which exploits SSD to accelerate distributed clients and improve the whole networks’traffic. |
PDF Full Text Request