Nimbus Data Systems Inc. claims its Gemini F400 all-flash storage array can support more than 4,000 simultaneous virtual desktop infrastructure users at less than $40 per desktop -- and it has a validated test report to prove it.
Nimbus benchmarked its array using Brocade Communications Systems' Gen5 16 Gbps Fibre Channel (FC) networking gear. The price per desktop only includes the storage array and not the FC switching, however.
Nimbus used the IOmark-VDI storage benchmark that the Evaluator Group created in 2011 to measure performance using real workloads. The purpose of the benchmark is to help IT users accurately assess storage performance for virtual desktop infrastructure (VDI) deployments.
According to the Evaluator Group's test validation report, the benchmark was achieved with a Nimbus Gemini dual-controller 2U F400 all-flash storage array with 24 TB of raw capacity. The array dedicated 17.6 TB usable capacity for the test, and featured 24 1 TB solid-state disks and a 4 TB cache with write-back caching.
Brocade contributed a 16 Gbps FC Gen5 6510 switch that connected the two servers running the IOmark-VDI testing application. The servers each had two, dual-port 16 Gbps Brocade 1860 host bus adapters (HBAs) for a total of eight ports.
The VMware View 64-bit Windows 7 virtual desktops ran on a VMware ESXi 5.1 hypervisor. The IOmark-VDI used storage clones placed on linked clones or fully provisioned desktops.
The tests determined the system could support 4,032 standard users or 2,016 heavy users before breaching performance thresholds. The array and the 24 1 TB solid-state drives used cost $160,000. That comes to $39.68 storage cost per desktop for standard users and $79.37 for heavy users.
IOmark-VDI tests capture I/O streams from running applications and then replay them to create workloads that are identical to real-world workloads.
IOmark's Nimbus Gemini and Brocade Gen5's system tests required all the storage needed for the real-world VDI to reside on the system under test, and 70% of the I/O response times could not exceed 30 milliseconds. Standard VDI users were defined as using 6.26 IOPS and .53 Mbps throughput, while heavy users were defined as needing 12.52 IOPS and 1.06 Mbps throughput.
Evaluator Group Senior Partner Russ Fellows performed the tests. "These were the best results we've seen," Fellows said. "We have tested and been involved with testing quite a few systems. This was the largest number [of supported users for one array] we have seen by a good margin."
Fellows said the most relevant result was that the system supported 4,032 uses at less than a five millisecond response time. The report revealed that 90.1% of the system response times were below 2 milliseconds; 99.4% of the response times were under 5 milliseconds; and 99.7% of the response times were under 30 milliseconds.
Fellows did add that it's not quite an apple-to-apple comparison to other storage systems that have run the benchmark. Nimbus was the first vendor to test using FC, and 16 Gbps FC is the fastest FC available. And the $160,000 price does not include switching and HBAs.
Nimbus CEO Tom Isakovich said flash vendors usually emphasize cost per gigabyte or cost per I/O in their marketing, "but that's just not really a very helpful metric when trying to make a decision about how to roll out VDI. Now we have crystalized what that cost really is."
He challenged other flash and hybrid vendors to run the benchmark.
"It's a very easy test for any vendor to run," he said. "This is an open benchmark. It took just a few weeks to run. Every storage vendor should run their hybrid array, or their all-flash array, through the paces on this benchmark because it's going to allow customers to make a very apples-to-apples comparison about what [an] array [is] capable of in terms of scale."
The IOmark testing puts storage systems through eight different replay workloads -- including Microsoft Office, Internet Explorer and Windows Media Server applications -- for standard and heavy VDI users. The benchmark sets specific thresholds to measure how many users the tested system can support within acceptable performance standards.