Hotness-aware Research On Reducing Write Amplification Of SSD And SSD RAID

With the emergence of massive amounts of data, the storage system faces more challenges than ever before. The wide adoption of Solid-state drives (SSDs) can improve the reliability and data access performance of storage systems, but SSDs also face deployment issues, in terms of the time-variant reliability, the limited erase operations and the cost of garbage collection (GC) operations, which may have a negative impact on the overall performance of storage systems. To reduce the influence, we take the character of workloads into consideration to conduct some research on SSDs and SSD-based RAID arrays, they can be described as follows:(1) Hotness-aware research on reducing write amplification of SSDsThe write amplification induced by garbage collection within SSDs extremely degrades the 10 performance and endurance of SSDs, and Real-world workloads generally exhibit high skewness in access patterns, which may aggravate the degradation. It is a consensus that separating hot and cold data may greatly improve SSD garbage collection performance. To achieve this, we first propose a light-weight and high-accuracy identification scheme, which is developed via a group of Least Recently Used (LRU) lists and requires only a small amount of memory and CPU cycles. Then we further deploy our scheme on SSDs with DiskSim simulator, and results show that with our hot data identification scheme, the GC cost of SSDs which is quantified by the number of additional page writes caused by GC is reduced by up to 73.1%under different workloads compared to the case of no identification. Even comparing to two state-of-the-art identification schemes, our scheme further reduces SSD GC cost by up to 59.1%(62.1%), and saves 44.3%(77.5%) of computational cost.(2) Hotness-aware research on reducing write amplification of SSD RAIDSSD-based RAID arrays can efficiently guarantee both the high reliability and high access performance of data, but the traditional parity update schemes with read-modify-write or read-reconstruct-write are not suitable for it. To address this issue, the recently proposed elastic striping scheme chooses to reconstruct new stripes with updated new data chunks instead of directly updating old parity chunks. However, it necessitates RAED-level garbage collection, which may incur a very high cost due to the mixture of hot and cold data. In this paper, we follow elastic striping and propose a workload-aware scheme to improve the performance and endurance of SSD RAID. In particular, we first develop a lightweight hot data identification scheme, then propose a hotness-aware caching approach to buffer incoming writes by categorizing data chunks into multiple groups according to their hotness values, and finally we propose a grouping-based elastic striping approach to separately write data chunks indifferent groups into SSDs. We also implement the proposed scheme on RAID-5 and RAID-6 arrays composed of commercial SSDs. Experimental results show that compared to original elastic striping, our scheme reduces 30.0%-70.6%(and 23.9%-63.2%) of chunk writes in the RAID-5 (and RAID-6) settings, and also reduces the average response time by 60.9%-79.3%(and 56.8%-80.9%) for the RAID-5 (and RAID-6) settings, respectively.