Small-size Request Optimization For Solid State Drives

Flash-based solid-state drives(SSDs)are gradually replacing traditional hard disk drives(HDDs)as the primary storage devices in many fields due to their impressive performance capabilities.Unlike HDDs,the minimum read/write unit of SSDs is the flash page,which is generally several times larger than the granularity of the I/O request sent by the host systems.As the size of flash pages continues to grow,the access unit mismatch problem between the host system and the underlying storage device has become increasingly noticeable.This mismatch problem leads the host system to send a large number of requests with request size smaller than the flash page size,which are referred to as small-size requests in this thesis.When processing small-size requests,whether the request logical address range spans two logical pages will have a negative impact on the SSDs for different reasons.Based on this observation,this thesis completed the following optimization works:Firstly,the across-page type small-size requests(across-page small-size requests)will be mapped into two flash pages in the address mapping process,resulting in additional flash read/write operations.To address this issue,this thesis proposes an across-page request remapping mechanism,which realizes differentiating mapping of across-page requests into a single flash page through a two-level mapping table.Subsequent access to the mapped request data can also be completed through a single flash read operation.Secondly,the small-size requests of non-across page type(intra-page small-size requests)will cause “partial write” to the flash page,thus resulting in a lot of page fragmentation.In view of this characteristic,this thesis proposes a partial programmingbased intra-page updating mechanism,which effectively alleviates the negative effects caused by intra-page small-size requests by writing the intra-page small-size request update data into the free space of the flash page where the original data is located,while avoiding the data interference problem caused by partial programming.Moreover,a data migration mechanism based on the update frequency of data is proposed to store the hot data with high update frequency in flash blocks with better endurance,thereby extending the lifetime of SSDs.Finally,in order to verify the effectiveness of the proposed mechanisms,the above two schemes and the comparison scheme are implemented on the SSDsim simulator.The experimental results show that the proposed optimization schemes for both types of smallsize requests can not only reduce the I/O response time,but also effectively improve the space utilization and lifetime of SSDs.