The Research And Implementation Of Adaptive Storage Strategy In Hybrid Storage System

With the flash memory technology keeps progressing, the capacity increases rapidlyand the price continues to decline, now Solid State Driver (SSD) based on flash memoryhas become a widely used persistent storage device. Compared with Hard Disk Driver(HDD), SSD has lower power consumption, anti-vibration and higher random accessspeed. However, SSD still cannot replace HDD to be mainstream storage device due tolimited reliability and high costs. But hybrid storage system that consisting of SSD andHDD gradually spread and has become a focus of current research.For the existing migration algorithm deficiencies of Cappella storage system, thispaper presents an object-based adaptive hybrid storage system AHSS (Adaptive HybridStorage Storage). In this model, the SSD and HDD are as the the same level of storagedevices, data are classified according to certain criteria, whereby complete the adaptiveplacement and migration between different devices. Using a method to identify thedocument that have critical impact on system performance, according to I/O requestaccess patterns, combined with number of access to the file and the relationship betweenfile size, the method classify files accurately, use priority list to maintain file accesshistory information, and set corresponding priority for the files. The proposed adaptivepriority-based data placement strategy and data migration strategy, can select theappropriate storage device for the files according to historical access information of filesand storage pressure of SSD. When files need to migrate, scan the priority list, thenmigrate lower-priority file first, if files have the same priority, migrate them according toLRU strategy.We used a variety of application workloads to test hybrid storage system, the resultshows that our hybrid storage system can divide file appropriately, the adaptive placementand migration strategy is effective for different types of loads, and has the versatility.Compared with the original systems based on ARC algorithm, under web services loadperformance increased by40%, under file server load performance increased by10%, andunder the mail server load performance improved by80%, even compared to full SSDstorage system, the performance gap is only about10%.