Comparing Cache-Accelerated Approach on DAS and NAS as Storage Solution for Small Scale Virtual Desktop Deployment – A Case Study at ABC University

Abstract

Desktop virtualization proposed a mechanism to consolidate desktop computing resources, where the processing power was moved from user's side to centralize server's side. Many of desktop virtualization initiatives considered as fail because it run smooth during trial, but facing bottlenecks during real-life operation. Not surprising, since during trial, we dealt only with very small quantity of VMs, compared with production. Some enterprise solution with focus on storage performance has proposed full-flash storage solution, but for small to medium scale organization that had budget constrains may prefer DAS (Direct Attached Storage) or NAS (Network Access Storage). Of course DAS and NAS had limitation in performance compared with SAN, then some vendors offer server-side caching solution for virtualization that utilize RAM (Random Access Memory) or flash drive to leverage the storage performance. This experimental test was conducted to evaluate DAS and NAS from two perspectives (Performance and initial-investment cost), using IOmeter as workload simulator and benchmark tools for five storage scenarios, all configured in RAID10 and using server-side caching as storage accelerator. This experimental test also intended to answer which storage approach (between DAS and NAS) that most suitable with ABC's requirement, ABC was a university located in Jakarta. Three scenarios from DAS-based approach, labeled as SAS7.2k, SAS10k, SAS15k, and two other scenarios from NAS-based approach was labeled as NAS1-7.2k, and NAS2-7.2k. When talked about performance, we focused on three critical parameters (IOPS, bandwidth and latency). While SAS15k (DAS-based) and NAS2-7.2k (NAS-based) were the top performer from each respective category, the author found that NAS2-7.2k was the best option, it outperformed SAS15k that 9,11% higher in price. NAS2-7.2k had 108,14% higher in total IOs per second (Represent IOPS), 108,1% higher in total MBs per second (Represent bandwidth), and 107,53% lower in average IOs response time (Represent latency) when compared with SAS15k.