VDI workloads have made virtual and storage admins look for storage technologies that have higher I/O rates and better performance. While solid state drives (SSD) can be a viable solution to increase storage performance as a whole, administrators still worry about TCO and ROI over the long run. This can ultimately lead to rejection of SSD and selection of cheaper mechanical disks that may not meet the VDI performance requirements and cause bottlenecks.
Matching physical desktop IOPS with mechanical disk is one of the biggest roadblocks in VDI deployment. Let us compare the IOPS performance and cost for different types of hard disk.
(TOSHIBA MK4001GRZB 400GB)
(Kingston SSDNow E100 400GB)
10K SAS Drive
(Western Digital Velocity Raptor)
Maximum Read Transfer Rate
Write Transfer Rate
Read - 90000 IOPS
Write –17000 IOPS
Read - 52000 IOPS
Write – 37000 IOPS
Target Storage and Server
Storage Read IOPS/Queue Depth
Storage Write IOPS/Queue Depth
The Price/GB comparison shows us the price of a mainstream SSD to be approximately 10 times the price of a mechanical hard disk. But the catch here is the read/write IOPS for SSD is 20 folds more than a mechanical hard disk. Hence using the right storage type for the right purpose is the key for a good storage investment.
For instance let us assume your enterprise has 100 clients on a host, and the hard disks are on RAID1 configurations.
- For Windows 7 clients to run smoothly the read/write ratio is 40/60 and the IOPS needed is 10 per VM
- Based on this for 100 VMs we need 1000 IOPS of which 400 is used for read and 600 for write.
Storage Cost Calculation (10K SAS drive in RAID1 vs. Enterprise Mainstream Performance SSD_
For RAID1 the IOPS is 50-60 for read, and 150-160 for write. That implies:
- (400/60) + (600/160) = 10.41 so that’s approx. 11 disks for 100 VMs
- 11 (disks) * 245 (price of 10K SAS) = $ 2,675 for hosting 100 VMs on 4,400 GB of storage
On the other hand, a single enterprise SSD can handle all the IOPS needed by the 100 VMs, costing about 3000USD having an IOPS rate of 90k for read, and 17k for write operations. But, on the flipside, you are left with 400GB of storage which is just 10% of what 10K SAS offers. For a larger environment that would use all more of the SSD IOPs capacity, a combination of SSD and SAS drive can provide the most cost effective combination of IOPs and storage space. By itself this makes SSDs more expensive, especially if you are looking for storage space and NOT performance. From there it is fairly simple math to calculate TCO and ROI for each hard disk type once you know your VDI infrastructure, the number of VMs and VDI usage and possibly the current IOPs identified with virtualization management software.
Storage administration in VDI is all about managing storage capacity and IOPS demand. Balancing this with keeping cost in mind is key for TCO and ROI. Otherwise, the whole purpose of deploying VDI to save IT cost may tend to fail.
A few points on how to maximize your storage performance and ROI are provided below.
- Use the Right Storage Device: To address this capacity vs. IOPS issue we can adapt the process of adding SSDs to handle I/O operations while using mechanical disks for back-end capacity
- Thin Provisioning to Save Storage: Thin provisioning of VMs on faster storage can save space on SSD hard disk. This allows more VMs per disk (but remember to monitor storage capacity as exceeding physical capacity can shut down your VMs).
- Memory Swapping: Having memory swapping occur on a SSD will allow applications that use swap memory to run faster while the actual processing is done on mechanical hard disks
- Segregate Based on Usage: You can segregate the usage of VM based on usage or departments. For example, for a department developing and testing VMs requiring faster I/O can run on a SSD while the marketing and sales that needs less IOPs performance can run on mechanical disk
- Moving the Snapshots: Non-active snapshots can be moved to a backup mechanical hard disk so it provides more average I/O for active snapshots
- Storage Tiering: Storage tiering is the process where the software determines the data based on type, usage and provisions it on mechanical or SSD disk automatically.
- Boot Storm Management: Boot storm occurs when all virtual desktops are turned on at the same time causing a spike on storage I/O usage. Timed boots or aligning storage arrays based on power on time can avoid boot storms.