Analysis of Block Access Address Correlation in Enterprise VDI Storage

摘要

Virtual desktop infrastructure (VDI) is widely used in enterprises and universities. Since the storage system is shared by many VDIs, it becomes a bottleneck. In order to solve this problem, we need to analyze and understand the I/O characteristics of real VDI storage workloads. However, there are no prior works analyzing I/O traces of large scale enterprise VDIs. In this work, we captured packets between servers and storage of an enterprise VDI service, and analyzed its storage trace. Based on accessed logical block address (LBA) and access time from daily traces, contiguous LBA were aggregated fixed-size blocks, and those blocks were then clustered according to their access frequency over time. The results showed that the same 50 GiB-sized clusters were found each day. And detailed analysis showed that only 26% of the LBAs were actually accessed for each of those clusters. Taking into account those block correlation results to improve storage tiering or caching seems a much promising direction to remove storage bottlenecks of current VDIs.