Internet shopping site Shopzilla recently turned to F5 Networks Inc.'s Acopia ARX file virtualization switches...
to get a better handle on more than 1 PB of files stored over three data centers.
File management is crucial for Shopzilla, which lists millions of products from thousands of retailers on its site and posts frequent updates based on seasonality and new products.
Burzin Engineer, Shopzilla's vice president of infrastructure, said he began testing file virtualization systems about a year ago, looking for global namespace and the ability to replicate data between different storage tiers from different vendors.
"I cannot have a half million to a million dollars of storage slowed down just because I want to virtualize it," Engineer said. "Acopia has no moving parts -- it is truly a virtualization switch. Its latency is almost negligible."
Engineer said his original intent was to manage and replicate data that is growing between 30% and 40% a year. He ended up with another benefit: flexibility in dealing with vendors.
"Primarily, I was hoping to gain a better handle on the rapid way data was growing," he said. "I try and use ILM to see if I can take data that's not used as often to my secondary or tertiary tier. The biggest advantage I got was I'm no longer beholden to a NAS vendor. I could now negotiate with other vendors. If I have a key partition of 7 TB residing on a particular vendor's storage, I can now grow that partition on another vendor's storage. I have the ability to stripe an existing volume across different systems, so I'm not beholden to one vendor."
The other major advantage comes from Acopia's Shadow Volume replication feature, which Engineer said copies files without impacting performance. Shopzilla's primary storage resides on systems from XIV Ltd., the clustered storage vendor IBM bought in January, and 3PARdata. It also has three tiers of NAS storage via ONStor gateways connected to XIV, 3PARdata and Nexsan disk arrays, plus BlueArc NAS boxes with its own set of disks.
Engineer said Acopia's ARX switches allow him to share files between his ONStor gateways and BlueArc NAS systems.
"The second big advantage is I can put production data on high-caliber storage," he said. "There was no easy way to replicate data in real time between production and development because one was on Tier 1 and one was on Tier 2. With Acopia in between, you can do cross-vendor replication. That was a huge win for us."
Engineer said out-of-band virtualization might be fine for migrating files, but he wanted an in-band switch to provide global namespace for all of his NAS files. "Having a layer that can be in front of NAS devices so you can take five different vendors and create one volume that exists across all the vendors' systems can only be done with inline virtualization," he said. "Out-of-band virtualization is an afterthought to me."
Engineer's one complaint about ARX is the way it stores the metadata used to apply policies to files. "Their metadata is a single point of failure, although they don't agree because they can rebuild it," he said. "It can take anywhere from five seconds to 10 minutes, but if that metadata volume goes down, you can have 10 minutes of downtime. I would like them to have redundant metadata volume. It's not a major thing, but it would be nice to have."