Research On Key Technologies Of Block-based Hybrid Storage

With the fast storage technical improvement over the last few years, Solid State Drivers(SSDs) are becoming a more and more important part of the computer storage hierarchy. SSD not only has the random access speed, the sequential access speed is also better than Hard Disk Drivers(HDDs) and its energy consumption is very low. However,the price and capacity of SSD limit its large-scale deployment. And hybrid storage is to be the choice. Hybrid storage is to get a single device consist of different storage devices by exploiting the each advantages to improve the system performance. Classifying data blocks and migrating data blocks are the critical problems in hybrid storage.In this thesis, through the I/O performance testing and analyzing on disk and solid state disk, found that the advantage of the random access performance of SSD is related to the request size, but not a simple linear relationship. To make full use of SSD is not simply placing the hot data blocks on SSD, but placing the critical data blocks which affect the disk performance most on SSD and taking the request size into consideration. Thesis proposes a kind of critical data identification strategy which based on access frequency,request size and seek distance. Compared to the traditional strategy which based on access frequency and request size, this method improves the identification of key data accuracy,places the data blocks which affect the disk performance most on SSD. As a result, the overall system performance would be improved. Meanwhile this thesis proposes a copies Migration strategy, which can effectively reduce the migration overhead and be well applied to the more reading and less writing working sets. Thesis implements this model at the device mapper level in the Linux kernel 2.6.31 to provide user a single device.Through the test, when the capacity of SSD is 20% of the working set, by using the new strategy, the execution time is reduced by approximately 23% compared to the HDD,which is about 2.3 times the SSD. When the capacity of SSD is 40% of the working set,the execution time is reduced by approximately 53% compared to the HDD, only 43%more than the SSD.