Flash devices have been widely used in embedded systems, laptop computers, and enterprise servers. However, the poor random writes have been an obstacle to running write-intensive DBMS applications on flash devices. In this paper, we exploit the recently discovered, efficient write patterns of flash devices to optimize the performance of DBMS applications. Specifically, motivated by a focused write pattern, we propose to write pages temporarily to a small, pre-allocated storage space on the flash device, called StableBuffer, instead of directly writing to their actual destinations. We then recognize and flush efficient write patterns of the buffer to achieve a better write performance. In contrast to prior logbased techniques, our StableBuffer solution does not require modifying the driver of flash devices and hence works well for commodity flash devices. We discuss the detailed design and implementation of the StableBuffer solution. Performance evaluation based on a TPC-C benchmark trace shows that StableBuffer improves the response time and throughput of write operations by a factor of 1.5−12, in comparison with a direct write-through strategy.

SSDs (Solid State Drives, or flash disks) have been considered as ideal storage for various data-intensive workloads, because of the low random access latency and the intra-disk multi-chip parallelism. However, due to inherent nature of flash memories, update-intensive workloads cause the flash disk fragmented, and trigger costly internal activities such as cleaning and wear leveling. We use database transaction processing as a motivating update-intensive workload. Our studies based on a flash disk simulator as well as flash disks show that, these activities result in significant overhead to the I/O response time and system throughput. To resolve the impact of internal activities, we propose dynamic page replications to exploit the multi-chip parallelism on the flash disk. Specifically, we replicate the frequently blocked data pages to improve the data availability even when internal activities block the request. To reduce the overhead of replications, we take advantage of the idle periods in the flash chips for the I/O operations by writes to replicas or reads from replicas, and further develop a prediction model for the decisions on those I/O operations to minimize the interference to normal I/O operations. We evaluate our techniques with three public transaction benchmarks in the simulator as well as on the real flash disks. Our results demonstrate the effectiveness of our replication management on improving I/O response time and system throughput.