All-Flash arrays from a variety of vendors are scoring major wins against traditional disk based storage arrays and as we’ve written to you before, flash storage seems poised to take over the datacenter as it is already present in almost every intelligent device layer in the solution stack. However where IOPS and throughput were once verifiable numbers that could be used to measure a variety of storage platforms against each other, there have always been ways to pad numbers with configurations and specialized workloads that would never see the light of day in a production datacenter and these propped up metrics have a singlular purpose of “proving” that Product A is better than Product B. The term I like for this is benchmarketing – a portmanteau of benchmark and marketing, in which the former is used in a misleading way to bolster the latter, possibly with statistical outliers or subtle inaccuracies.
It’s well known that flash is changing what’s possible in the realm of storage performance, and common benchmarking tools like IOmeter can be made to display very gaudy numbers if the right workload is presented to All-flash arrays. This is especially true during Proof of Concept exercises where there the array is mostly empty and there is no contention from other workloads. Marketing guys tend to love these numbers, but they simply aren’t attainable with real world workloads with many of these systems as configured. To address this issue the Storage Performance Council (www.storageperformance.org) offers standard workload configurations that all competitors may run against and post public results for verification, related to IOPS and IOPS/$ based on the list price of the configuration. Unfortunately not every vendor submits a public benchmark, usually because it would conflict with their benchmarketing number. If a standard real world benchmark workload is not applied then how can you be confident the IOPS numbers they claim will hold up under load? More importantly, why doesn’t every vendor supply a public benchmark to begin with? What are they hiding?
In an attempt to make All-Flash arrays a cost effective option, a means to store more data on less physical media had to be devised in order to bring cost on par with legacy spinning disk. Unfortunately benchmarketing rears its head here too. As new storage software is being written that is optimized for flash we are seeing deduplication and compression ratios that were previously unheard of for active file systems, with vendors claiming effective capacities that are 4-6x the raw capacity of the array. While certainly achievable in some use cases, not all data deduplicates well and every customer environment is different. And what happens when the array gets busy? Does deduplication continue to work or does it downgrade from in-line to post-process when the array can’t keep up? Don’t be surprised if you don’t get a murky answer from your EMC sales engineer when you ask. To avoid this, when evaluating All-Flash arrays, ask your vendor for a written guarantee for the useable storage capacity you will get from the array. EMC won’t give you one for its XtremIO platform. Neither will Pure Storage. Why is that? It’s important to look past the gaudy numbers when making a purchase for something as critical as enterprise storage and we recommend considering storage vendors who submit public benchmarks and place written guarantees on useable capacity.
That is why Tego Data has partnered with Kaminario (www.kaminario.com), a Massachusetts based All-flash enterprise storage company founded in 2008, who until last week held the #1 and #2 spots on the Storage Performance Council SPC-1 benchmark. Kaminario arrays now hold the #2 and #3 spots, unseated by Hitachi’s latest offering. Unlike other All-flash array vendors, Kaminario K2 arrays deliver consistent IOPS and sub millisecond latency, even under 100% load. Additionally, their unique DRAM caching approach means that an entry-level 24 SSD K2 array is already capable of maximum performance without adding more disks, with truly linear performance scale-out being achieved by connecting additional active-active K2 arrays together which are then treated as a single array allowing mixed controller and SSD capacities. Kaminario also does what few other All-flash vendors will, by placing a written guarantee on the useable capacity you will get from a K2 system. The entry level Kaminario K2 starts at 9.6 TB of raw flash (24×400 GB SSD) and guarantees end users will see 22-44 TB of useable capacity, depending on how efficiently their dataset can be deduplicated. Kaminario backs up their guarantee by giving customers free hardware until the guaranteed useable capacity reflected on the quote is met, and every Kaminario array comes with a similar guarantee. Further breaking the mold, Kaminario arrays have all-inclusive software licensing, and when new features are rolled out customers receive them for free as part of fully automatic non-disruptive upgrades.