One company's move to go virtual leads to I/O constraints

Replicating data in a virtual machine means you can be storage system agnostic

Austin Radiological Association is a quarter of the way toward virtualizing its server and desktop infrastructure, but the move has placed a strain on its networked storage and slowed application performance.

ARA, which provides diagnostic imaging services through 15 outpatient centers in central Texas, is trying out a new product from startup Zerto, which launched last year with an unusual approach to data protection. Instead of protecting data by replicating it offsite from storage arrays, it uses software to replicate the data within the virtual machines.

[ Keep up on the day's tech news headlines with InfoWorld's Today's Headlines: Wrap Up newsletter. ]

ARA's IT department has about 40 people managing almost 800TB of data. It has about 1,000 desktops for 800 employees along with 300 servers, about 25 percent of which are currently virtualized using VMware and Citrix XenServer.

ARA, which is part of a larger consortium of radiological imaging firms, had originally imagined itself moving to a private cloud, carving up its data center, selling services to its regional affiliates and becoming the "center of the universe." But that became complicated fast.

"We backed off that cloud marketing hype and decided to learn to crawl before we sprinted," said ARA CIO Todd Thomas. "We didn't see ourselves doing chargeback right away or auto provisioning of servers right away, because our processes weren't mature enough as an IT organization."

So ARA shifted its plan to virtualizing its data center, selecting Cisco's Unified Computing System (UCS) blade center as its server farm, and EMC and NetApp products for storage. But what Thomas didn't want to do is get locked into a particular vendor's data replication software.

"If I want to switch colocation providers, it's easier to migrate a virtual machine and not care about what the underlying storage platform is," Thomas said.

As with most companies, ARA's data replication scheme for business continuity and disaster recovery is based in its networked storage, which features both EMC Symmetrix DMX-4 950 and NetApp 3140 arrays. Replication in those arrays is performed at the block or file level, meaning data is either mirrored or copied to its offsite recovery facility 20 miles south of its primary data center in Austin. ARA uses both EMC's Symmetrix Remote Data Facility/Asynchronous (SRDF/A) and NetApp's MirrorView software for data replication.

But when it began virtualizing its server and desktop environment, the strain of additional virtual machines overwhelmed its storage arrays' I/O bandwidth.

The ARA project started as a XenApp installation in which applications were going to be streamed down to wise desktop terminals. "In that rollout, we ran into a number of hurdles that shifted our strategy to go more the VDI route than just the applications streaming route," Thomas said.

But an "unforeseen" explosion of virtual desktops created I/O performance problems. "We got about 70 percent of the way through the project and are now on hold," he said. Instead of 80 percent streaming applications and 20 percent VDI, ARA wound up implementing 80 percent virtual desktops and 20 percent streaming apps.

"The predominance of virtual desktops created a strain on our storage arrays we weren't anticipating," Thomas said. Not surprisingly, application performance began to suffer.

"We got emails from our customers saying, 'We can't work,'" said Jack Yudell, ARA's data center operations manager.

So Thomas and Yudell decided to move toward server-based data replication.

That's where Zerto came in. Zerto's Virtual Replication software performs asynchronous replication of data at the virtual machine disk (VMDK) level rather than at the storage volume level.

The Zerto Virtual Replication product features two software components: the Zerto Management Server and the Zerto Virtual Appliance. The Zerto Management Server resides on the VMware vCenter Server as a plug-in to the vCenter Client. A Zerto Linux-based Virtual Appliance sits on each ESX host for which it's replicating virtual machines. Each Virtual Appliance is installed on the ESX host by an automated process from the Management Server.

The Zerto Virtual Appliance software is deployed on top of VMware's vSphere cloud software. Zerto's software looks at the I/O stream associated with a given virtual machine (either a VMDK or VMware Raw Device Mapping) and continuously replicates data from user-selected VMs, compressing and sending that data to the remote site over WAN links. Each Virtual Appliance handles its own load and replication process and there is no need for any data movement between ESX hosts.

"We are agnostic to storage; you can have a combination of vendor arrays, and we can replicate to another third-party array," said Gil Levonai, vice president of products at Zerto.

Laura DuBois, an analyst at market research firm IDC, said Zerto's replication product created a new market segment.

"Zerto leverages continuous replication in concert with a checkpoint process, rather than array or VM-level snapshots. This allows for point-in-time recovery from checkpoints, which are taken every few seconds. Thus organizations can use Zerto not only for business continuity and disaster recovery but also for recovery from logical errors and corruption," DuBois wrote in a white paper.

Zerto claims to have become the first vendor to launch a hypervisor-based data replication product when it shipped its product in June. VMware followed suit in August, with its vSphere SRM 5.0 data replication feature. With the Zerto Virtual Replication product, asynchronous replication occurs at the VMDK level rather than at the storage LUN level.

According to Zerto, it offers the same scalability, performance levels, recovery time objectives and recovery point objectives found in array-based replication products. The product includes WAN optimization and compression, automated failover and failback, and signature mapping to re-sync changed blocks during a WAN recovery.

ARA tested a beta version of Zerto's software earlier this year, and both Thomas and Yudell were impressed. "It did everything the guys said it would," Thomas said.

They plan to have ARA's virtualization project, including installation of Zerto's product, completed within a year.

"So VMware allows us to decouple hardware from software," Thomas said. "And Verto's replication technology does that for our replication strategy."

Lucas Mearian covers storage, disaster recovery and business continuity, financial services infrastructure and healthcare IT for Computerworld. Follow Lucas on Twitter at @lucasmearian, or subscribe to Lucas's RSS feed. His email address is lmearian@computerworld.com.

Read more about virtualization in Computerworld's Virtualization Topic Center.

This story, "One company's move to go virtual leads to I/O constraints" was originally published by Computerworld.

From CIO: 8 Free Online Courses to Grow Your Tech Skills
Join the discussion
Be the first to comment on this article. Our Commenting Policies