Design And Implementation Of A Multi-user Shared SSD Cache Algorithm

In recent years, virtualization technology becomes more and more popular. In this situation, each user has a separate logical space, but they share the public physical resources. With the increase of CPU computing speed and network bandwidth, the storage system becomes the bottleneck of the system. Because SSD(Solid State Drives) has the advantage of high performance and low power consumption, storage system can improve the storage performance by using SSD as cache. This hybrid storage system has gradually become a research focus in the field of storage.Algorithm in this paper builds a cache storage system by using SSD and with the use of RAM as a disk cache layer. In the face of multi user application scenarios and system acceleration requirements, the cache layer is divided into sequential area and random area logically. Accordingly, this system comprises a sequential request identification module and a random hot request identification module. The sequential area uses RAM completely. The sequential request identification module is used to identify the sequential requests which are usually randomized in the multi user situation,and then sequentially pre fetch the data to migrate to disk, so that it can reduce the performance loss caused by the sequential requests which are randomized in the disk.The random area is composed of RAM and SSD which are divided into read area and write area, respectively. In the random hot request identification module, heat value of a data block is defined to be related to the user priority. It is preferred to meet the high priority user's cache needs. According to the heat value and the reading or writing attribute, it identifies the hot data block or secondary hot data block which is mainly read or written, and then moves blocks to RAM/SSD's read/write area according to the strategy. According to the difference of the heat threshold value of the reading/writing area, the read/write area ratio is adjusted dynamically. As a result, this module speeds up the speed of frequent access, at the same time, appropriately reduces the number of writing SSD. These two modules constitute the upper and lower level relations. Through the filtering function of modules, it reduces the amount of metadata during identifying random hot data blocks.According to the design scheme, the system is implemented and tested. The test shows that, the design scheme can identify the sequential requests of different users. To deal with the small requests, there can be several times to improve the performance of processing, and for large requests, the design scheme can also accelerate the function of 10~20%. For the processing of random requests, the design scheme significantly reduces the cache pollution allocation.Compared to the LRU scheme, it has a substantial performance improvement.