Senior Systems Engineer
Data Storage Services
Mainline Information Systems
Moshe Yanai zigs where others zag. Gil Press, writing in a recent Forbes article describes Moshe’s 40 years of re-inventing storage. In the article, Gil distills the five rules for Disruptive Innovation, as he describes Moshe’s formula for his (and his team’s) next storage creation.
For this post, we are interested in Rule #1 from that article:
Study the conventional wisdom and find out exactly what’s wrong with it.
As mentioned in my initial post: Infinidat’s Preeminent Caching, it was stated that SATA is crucial to the overall design. (The Infinibox contains NL-SAS drives. SAS protocol adds command queueing and additional concurrency to SATA platters. For performance sake, we are concentrating on the fact that these are 7200 RPM SATA-based HDD.) But, how or why is NL-SAS crucial? It is crucial to be able to put Petabytes of storage on the floor far cheaper than the competition. Chip Elmblad at Infinidat speaks to several advantages of NL-SAS as backing store here in a blog post. Check out the charts Chip references that show the cost differentials: SSD vs. NL-SAS. I had intended to bring up several points that Chip wrote about, but he beat me to it – and thankfully, that allows me to shorten this post. His post is a good read, and I’m echoing several of those themes.
Today, depending on what your sourcing is, you will find that Enterprise SSD costs vs. NL-SAS is 13 to 17 times higher, per GByte. Higher demand for flash across a broad spectrum (phones, tablets, storage) introduced increased costs in the second half of 2016. Consequently, flash costs are higher in 2017 than they were in 1H 2016, with added spot shortages showing up today. In 2016, the Gartners and IDCs of the world had forecasted that by now, Q4 2017, the shortages would ease. However, recent analysis shows that flash shortages and subsequent higher costs (compared to first half 2016) are to continue, at least, through mid-2018. Additionally, it has been projected, by industry experts (including Western Digital), that Enterprise SSD costs will stay 8 to 10 times higher, per GByte, than SATA, for at least a decade.
As the above for backdrop and returning to Rule #1, what exactly is wrong with conventional wisdom as it applies here? Conventional wisdom dictates, to deliver a high performing Enterprise storage subsystem, it is a must that the backing store be flash. Infinidat turned that on its head by designing a system that performs IO out of cache (see that blog post link above as to how that is accomplished).
Repeat an Infinidat mantra after me:
- Cache is for IO
- Storage is for Data
If 95% of your IO is out of DRAM (see details in initial post), your backing storage simply needs to be able to deliver bandwidth, as most of the back-end IO in an Infinibox is sequential write or sequential read (thanks to log-based writes and careful positioning of similar data). NL-SAS (SATA) is quite good at delivering decent bandwidth per hard drive.
Additionally, it would have made little or no sense to design an Infinibox with anything OTHER than SATA as the backend, especially if the design simply requires modestly performing bandwidth in and out of the backend. Just add more bandwidth (additional HDDs) to hit the desired design center.
But there’s more!
With NL-SAS as the backend, as a customer, you can put more high-performing Infinidat storage on the floor at a cheaper cost than other solutions. Infinidat claims they are profitable when many startups aren’t. Several storage startups were purchased, just in the nick of time, prior to making very difficult business decisions regarding their future – others weren’t so fortunate.
Currently, with at least a 10x cost advantage SATA vs. flash, there is no reason to doubt that an Infinidat solution has a considerable pricing advantage, come deal time.
Anecdotally, a recent purchaser passed on that the Infinibox solution they purchased, was one-half the cost of a competitor’s solution. After factoring in the two-to-one compression, common in an Infinibox, they are today enjoying one-fourth the effective cost over the competitor’s solution.
A few closing thoughts, to counter objections I hear on occasion. Suppose, in some alternate Infinidat business/engineering-challenged universe, instead of Faster/Cheaper/More Reliable design criteria, the decision was to go with a Faster/More Reliable design. That resulting product would have joined a slew of very similar established products and new storage solutions. This twilight-zone Infinibox would have been much less of a differentiator – disrupting nothing in the process – upsetting their raison d’être. Another objection is, what if the cost differential NL-SAS vs. Enterprise SSD declines to 8x in 18 months and is trending downward – doesn’t that prove they should have gone with flash out of the gate? I suppose they and their partners make less margin, and Infinidat researches if the SATA->SSD crossover is a few years out, and prepares for that.
Please contact your Mainline Account Executive directly, or click here to contact us with any questions.