Performance Evaluation of NFS-based Primary Storage with Deduplication using Windows Server and RAM-based Cache on Small-scale VMware Environment

Deduplication in storage allows us to eliminate duplication in order to maximize the usage of storage capacity. The majority use of deduplication was for backup functions, but also applicable for primary storage. One of good candidate for deduplication implementation is desktop virtualization. Deduplication is resource intensive, so its application to the primary-storage can have serious implications on overall system performance that negatively impact user experience. This paper attempts to perform performance evaluation by applying deduplication (post-dedupe approach) to the primary storage, the implementation is done by creating NFS-based virtual storage (vstorage) using Windows Server 2016. The use of vstorage itself has a significant impact on storage performance, the number of IOs and bandwidth reduced to 13.33 times when compared to the basic performance of storage without vstorage. Under these conditions, it is strongly recommended to use caching mechanisms, in the context of this paper, the author used RAM as a red-and-write cache. The use of cache proven to improve performance, even faster than basic storage performance up to 1.74 times. Furthermore, the author also performed performance test while the dedupe run, causing decrease in performance up to 4.84 times compared to the performance when the dedupe process is not running. The use of Windows Server 2016 as a deduplication solution overcomes storage capacity limitations, and the use of RAM as a red-and-write cache becomes the solution for storage performance. Furthermore, the author creates 2 vstorages on an existing server that has 35 VMs, where the average number of IOs per VM is 401.94 IOs for vstorage #1 (13 VMs) and 378.24 IOs for vstorage #2 (12 VMs). The author also tested this configuration for 4 months and got positive feedback from 93,33% (14 of 15) of the respondents.