3dmentat - Fotolia

University uses Tintri storage to tame 'bad-behaving' VMs

The University of California Irvine found Tintri VM-aware storage to improve performance and analytics for its virtualized workloads.

When its legacy storage system began struggling to keep up with more than 700 VMware virtual machines, the University of California Irvine (UCI) looked at a wide range of replacement options. The university's IT considered all-flash and hybrid arrays, hyper-converged and object storage, and storage designed for virtual machines.

UCI picked Tintri Inc. as its storage for virtualized workloads. John Ward, IT enterprise architect at UCI, cited Tintri's performance, analytics and integration with VMware vSphere as its main advantages over the competition.

Ward said he started thinking about a storage refresh last summer when his NetApp FAS3240 started slowing down. He said he continuously evaluates storage vendors to see if any interesting new technologies arise. "We always keep an eye on them," Ward said.

Ward said he looked at most of the major storage systems and all of the startups he found interesting. The final five came down to NetApp Inc. and smaller companies, such as all-flash startup Pure Storage Inc., Nimble Storage, Tegile Inc. and Tintri. Ward sent 92-question surveys to those five and received price quotes.

In March, UCI selected Tintri and installed two VMstore T880 arrays -- the largest of Tintri's three models. Each hybrid T880 has 78 TB of usable capacity, with 8.8 TB of flash.

Ward said he also considered hyper-converged and object storage systems before whittling his list to five, but didn't consider those technologies the best fit for his storage.

"We were looking at all types of technology," Ward said. "We even looked at some software applications to put in front of your storage to improve performance."

So far, Ward's team has moved approximately 80% of its virtual machines to Tintri storage.

Flash performance, troubleshooting make Tintri storage the best bet

He said the biggest problem Tintri storage solved was performance, although it brought other benefits.

"It seems like a lot of vendors have speed," Ward said. "One of the other problems we had was being able to determine which VM was causing [problems]. Tintri's performance metrics are very good, so when you do have an event or a machine is behaving badly, it's easy to track down which one it is."

As an example, he said UCI had a "fun event" soon after installing Tintri. "One morning, we came in and a machine was polling 3 gigabits of the 10-gigabit network that was available," he said. "One machine was polling all that bandwidth, behaving badly. It only took three clicks to see which machine was doing that.

"Everything else was seeing under two-millisecond latency on storage, so that was good. That kind of event would have thrown our older hardware into a bad state, and we would have had a problem identifying which machine was doing it. We could also go take a bad-behaving machine and look at its I/O [on Tintri storage]."

He said another thing he likes about Tintri storage is "you don't have to do anything to it. We set it up and it runs. We don't have to spend any staff time to make it happy. I'm happy with how little time we have to spend doing anything with the device."

Much of Tintri's performance comes from the way it uses flash in its hybrid arrays.

"Tintri calls their process FlashFirst -- everything coming into the Tintri gets put on flash, and things that aren't accessed as much get pushed out to SATA drives," he said. "One concern with that is, what if your working set is bigger than the amount of flash? I suppose the Tintris would have problems if we got to the place where the working set is bigger than the flash [cache], but that's not different than any system that's not all-flash.

"NetApp had some early all-flash, but it was extremely expensive at the time for an all-flash FAS. Their hybrid arrays were also expensive, although in the ballpark of everybody else's. But NetApp doesn't have really good insights into what's going on in terms of performance."

Ward said his biggest complaint with Tintri's VM-aware architecture is more of a VMware issue than a storage problem. He would like to see VMware allow the array to store snapshots initially, rather than having vSphere initiate the snapshot and then move it to the array.

He would also like to see better integration between Tintri and his Veeam backup software, which was also developed specifically for VMs. Veeam software supports array-based snapshots for NetApp and other large storage vendors, but not yet for Tintri.

"Veeam is aware of copies of data preserved in snapshots with NetApp, but it is not capable of doing that with the Tintris," Ward said. "But Tintri has the APIs that Veeam needs to make it happen, so it's of matter of when Tintri will become a high enough priority with Veeam."

The next storage step for UCI will be to migrate the rest of its data off the aging NetApp arrays. The data that is not vSphere-centric won't go on Tintri. Ward said new NetApp arrays are a possibility, but he is exploring other options.

"Our Windows team would like to move from CIFS on NetApp to having it done by a Windows File Server," he said. "Windows does file caching, and there are other things you can do with the Windows File Server that you can't do with CIFS."

Next Steps

New Tintri VMstore model moves it to the enterprise

Tintri's operating system upgrade brings better QoS, snapshotting

Housing provider replaces EMC with Tintri hybrid array

Dig Deeper on Hybrid flash arrays