Tintri designs its arrays from the ground up for virtualization and flash. Plus, the units of management and IO allocation on Tintri arrays are per-VM, so Tintri’s architecture doesn’t let a single workload crowd out others. But this design makes testing harder, because the system is designed to allocate IO fairly to workloads, rather than letting one workload take over and impact others.
Fortunately, Trevor Pott recently wrote a very detailed, in-depth review of the T850 Hybrid-Flash Array in The Register. His testing methodology is a breath of fresh air—rather than using irrelevant hero numbers, which can damage credibility by setting up unrealistic expectations, Trevor tested in real-world production environments.
Trevor captured what customers really care about–testing and validating vendors’ claims in real-world scenarios, over a long period of time. Today, we’ll discuss key takeaways from Trevor’s blog.
Trevor tested Tintri’s middle-of-the-line model among the three available for the T800 Hybrid-Flash Array family.
He observed double-digit space savings with VDI and 1.9x with a real-world workload, where uncompressible JPEGs consumed most of the space. Once again, these weren’t just “synthetic benchmarks,” which Trevor says aren’t good enough for “rational recommendation.” This is Tintri in a production environment.
Trevor achieved approximately two gibibytes per second throughput, which is pretty close to theoretical maximum. But more importantly, his real-world workload saw the Tintri running between 17,554 IOPS at 1,116 mebibytes per second throughput and 20,084 IOPS at 1,264 mebibytes per second throughput.
Note that this is with workloads using multiple different block sizes and constantly shifting amounts of read/write balance and randomness. As Trevor says:
Delivering 20K IOPS-ish whilst serving up roughly a gibibyte of data per second is a pretty decent array in my books. These figures aren't marketing voodoo, either. They're the performance you can actually expect when you are running multiple workloads in a mixed workload environment.
Finally, no testing of a Hybrid Array is complete without testing how hot and cold data is treated and how performance drops when flash limits are reached. Trevor observed that Tintri Hybrid-Flash units are good enough to go up against real-world all-flash arrays for most workloads.
Simplicity is at the heart of Tintri’s design, and the UI makes it easy not only to track real-time analytics but also to get to the bottom of any problems to isolate errors. As Trevor says:
“Spikes in latency, dips in flash hit ratio, or any other anomalies can be drilled into to see where the fault lies. Was the network being slow? Was the latency due to a flash miss, or did the flash miss occur and latency was mostly unaffected?”
Trevor effectively captures the essence of Tintri design. It’s simple, it’s VM-aware, it’s set-and-forget—and that’s how a storage array should be. And as soon as Trevor's article published, our engineering team picked up some of Trevor's recommendations, submitting them to the development cycle immediately. Now that's fast response time.
Even better, Trevor didn’t focus on just lab tests, but a combination of lab and real-world testing. And while you're at it, check out our T5000 All-Flash Array, not to mention Tintri OS 4.1, Global Center 2.1, and VMstack. We're offering real change in the market, and we're gratified that Trevor and our customers are right behind us.
I'll let Trevor finish this one:
“The Tintri T850 I've had to work with is good enough to run business workloads. From SQL to Exchange to VDI to image rendering, and all of these at the same time. It degrades gracefully when at the redline. Years of testing storage solutions have led me to believe this is a reasonably rare combination of features. Frequently advertised, rarely delivered.”
Unique control with VM-level actions for infrastructure functions including snapshots, replication and QoS make protection and performance certain in production, and accelerate test and development cycles.