IO Scheduling Optimization Based On The FT Processor Storage System

With the calculation speed of processing chip increasing, storage technology has become the bottleneck for the further development of computer system, storage wall has become difficult barriers of computer system performance to break through. Therefore, the problem of efficiency of the storage system has become a hot and difficult problem in computer Field. SSD with its inherent stability, efficiency and many other advantages has become the best choice of computer system to further improve the storage efficiency. In computer system software, file system layer, common block layer and I/O scheduler layer are integrated into a number of optimization programs for mechanical drives. Simplely replace computer mechanical hard disk solid state disk storage system, which is not only difficult to play outstanding performance of SSD, but unpredictable consequences will appear. Therefore, how to effectively solve the problem of solid-state disk storage systems matching with the current system software is the key to efficient use of solid state disk storage systems, but also the current research hotspot for SSD.There is still a gap in processing frequency between domestic FT processors and mainstream commercial processors, but FT processor has realized multi- core and multi line parallel processing at lower working frequencies. Multi core multi line parallel processing method reduces the single core calculation frequency, divided equally the computational resources for multiple computing core. For the application scenarios with high concurrency scenarios and low computational requirements, FT processor with low single-core performance can effectively meet the computing needs but also will not bring a substantial increase in power consumption. FT processor is especially suitable for the server with high concurrency and much I/O consumption, but at present, problems of domestic storage system based on FT processor still exists as follows:(1)SSD has an excellent internal concurrency mechanism, which is from afar multi-core parallel processing of FT processor. But it is difficult to realize the parallel I/O due to the funnel design of traditional system software.(2)Under the massive data environment, data distribution has showed certain rules, and it can deeply affect processing per formance of the system. At present there is no corresponding adjustment to make according to the data characteristics.(3) SSD inherent asymmetry defect of read and write exists, there is one to two orders of magnitude gap between read and write performanc e, this defect has seriously affected the stability of the I/O path. Therefore, according to the problems existing in the current system software, this paper has carried out researches as follows:First, according to the multi-core and multi line parallel processing characteristics of FT series processor, we have designed and realized new-type software structure of parallel I/O scheduling. The new-type scheduling software has two levels of front end and back end. Based on FT multi-core, the front end has designed parallel scheduling mechanism to realize fair scheduling of multi nucleus processor. According to the distribution characteristics of data and performance deficiency of SSD, the back end has respectively realized different optimization strategies.Second, for the problem of system showed sudden localized congestion and higher average response delay time with the application layer heavy-tailed distribution data, this paper proposes DVIOS scheduler. DVIOS scheduler can be controlled according to the subsequent feedback dynamic to regulator valve, according to the classification of the real-time feedback dynamic to adjustment request for the upper I/O, so as to adjust real-time local congestion of system I/O path. The experimental results show that, the DVIOS scheduler can effectively improve local congestion of system I/O path and significantly optimize the system average response delay time.Third, although the overall performance of SSD is superior, read and write asymmetry defects still exist, NOOP scheduling mechanism is difficult to remedy this defect. This paper puts forward the HTIOS scheduler, classifying the upper I/O requests according to the type and size of request to realize the differentiation scheduling of the read and write requests, to make up for the read and write performance asymmetry defects of SSD. Experimental results show that the HTIOS can effectively optimize average response delay time of the storage system and stabilize the overall performance of the system.