Sizing VDI storage for I/O performance in flash arrays

Feature

Sizing VDI storage for I/O performance in flash arrays

Virtual desktop infrastructure (VDI) is an increasingly popular choice for user workstations. VDI provides better control over desktops, allows user data to be stored centrally, reduces support overheads and cuts hardware maintenance and refresh costs.

But key to its implementation is VDI storage that can handle the IOPS demands of virtual desktops, which are fairly low – though random – during the working day, but which spike at the beginning of shifts as staff log on in so-called boot storms.

VDIstorage.jpg

VDI storage requirements are hard to satisfy using spinning disks without using short-stroked spindles that waste capacity and energy. So, many select flash-only or hybrid flash disks as VDI storage, which have the advantage of providing much greater I/O performance than just spinning disk and come in a variety of form factors.

But how do you size a VDI storage deployment to meet the performance needs of your desktop environment?

This article looks at two customers that took two differing flash approaches to VDI storage – one hybrid flash with spinning disk, the other pure flash storage – and how they determined the IOPS requirements they needed to fulfil. 

Violin's pure flash powers Charityshare's VDI storage

Charityshare is a not-for-profit consortium providing managed IT services to around 5,000 users at Age UK, The Children's Society and Alzheimer's Society.

"Alzheimer’s Society and The Children's Society have been running WinXP VDI on an HP EVA SAN for the last two or three years," says Greg Cooke, engineering manager at Charityshare. "When Age UK joined us, it wanted to refresh its desktops."

So Cooke and his colleagues looked at the charity's needs, undertook staff surveys, calculated requirements, and found that 80% to 90% were suited to VDI.

"We had to come up with some approximate figures of what was required in terms of IOPS and capacity," says the organisation's Citrix engineer, Paul Mainstone. 

The decision was down to usability, cost and support. We chose Violin Memory because of its comprehensive support

Greg Cooke, Charityshare

"In a VDI environment there are times of day when I/O is more important, such as the start and end of the day. We had to make some assumptions around boot storms and logon storms, and consider how anti-virus would affect our storage as this can have a big impact.

"We've found in the past that sharing storage between servers and desktops has an uncertain effect on user experience," he says. 

Charityshare has found that VDI is more sensitive to disk performance than a server, so a large SQL job running on a shared disk causes desktops to lose performance as applications lock up and freeze, while server applications can deal with reduced I/O much better.

The consortium knew it needed to find a storage solution that would deliver the IOPS and capacity while remaining within the charities' budgets. It shortlisted three suppliers, all of which delivered the right features. 

"In the end, the decision was down to usability, cost and support," says Cooke. "We chose Violin Memory because of its comprehensive support and backing from HP at the time.

"We decided on the cheaper MLC flash storage, largely because we sized the desktops at around 35 IOPS/desktop, which was always going to be 1.5 times more than we actually needed," says Mainstone. "The overall requirement was for 70,000 IOPS at boot storm and 8TB of capacity. Violin's system delivers more than 100,000 IOPS, so we knew it would more than meet our technical needs and be within budget.

"The implementation process was straightforward and painless," says Cooke. "Violin were great. They installed an evaluation unit in our datacentre and we were up and running the same day. We load-tested several hundred desktops and found it delivered what was required."

After deployment into the production environment, Charityshare found it was using fewer IOPS than expected. "As part of the feasibility study, when we classified our user base we were over-generous in our calculations of the IOPS required for task and knowledge workers. This means we have more capacity for the future if requirements increase," says Mainstone.

"Now it's installed, from the start of roll-out to the end of roll-out, IOPS performance has not declined and there have been no outages, so we're very happy with the performance. The proof of the pudding though is how users respond, and they have been extremely positive."

Nimble hybrid VDI storage delivers IOPS for Saunderson House

Saunderson House is a firm of independent financial advisers with around 100 users and infrastructure based on VMware ESXi and VMware View 5.1. The company first used VDI at its disaster recovery location, and found it worked so well it migrated all users to virtual desktops.

We looked for flexibility, adaptability and low step costs. Nimble means we don't have to replace the whole unit, but we can expand it as necessary

David Pritt, Saunderson House

"I have used VDI for two-and-a-half years," says business systems manager David Pritt. "I found it stable and easy to manage. Now we've made the decision to go completely VDI. This gives us more flexibility, means staff can work from home or from another office if necessary, supports BYOD [bring your own device], and it's easy to manage."

VDI means Pritt can simplify his desktop estate of 110 PCs, each of which differs due to user customisation. "It's not easy to lock the PCs down due to the specialised software we use," he says, "however, View 5 works off a set of gold images, one for each task, and linked clones.”

"We run two Hitachi AMS 2100 SANs – 12TB here, 8TB at the disaster recovery site – and they're solid and work well," says Pritt. "But when we started VDI storage load-testing, we realised we saturated the controllers quickly, especially at logon time when we experienced boot storms. Hitachi said we needed to buy another tray with more spindles, but I didn't need the extra capacity, I needed more IOPS."

Pritt then looked at flash-only storage suppliers, including Tintri and Pure Storage, but preferred the hybrid option offered by Nimble as it incurred no tiering management overhead.

"I also liked the idea of a hybrid system rather than pure flash because we could store more data than on pure flash," he says. "I still have reservations about flash, as it degrades and so needs clever algorithms to avoid noise between cells.

"I needed to be convinced," he says. "Nimble Storage sent us a box on approval, so if we liked it we could keep it. In the end, we bought two CS210s with about 8TB each and we use them just for the VMware View environment. We tested with 8,000 IOPS and the Nimble storage wasn't even stressed, whereas the Hitachis would have been saturating."

Pritt says the system offers replication, snapshotting and compression, and he is impressed by the single pane of glass management system. He also liked the Nimble system's expandability.

"We looked for flexibility, adaptability and low step costs," he says. "Nimble means we don't have to replace the whole unit, but we can expand it as necessary. It offers more flexibility than products like NetApp, where you hit the top then you've got a major investment step.”


Email Alerts

Register now to receive ComputerWeekly.com IT-related news, guides and more, delivered to your inbox.
By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

This was first published in April 2013

 

COMMENTS powered by Disqus  //  Commenting policy