Over the past several weeks, I have been on a very busy analyst briefing tour. All very smart, articulate and pleasant people. Talking with them is one of my favorite activities I get to do here at Tegile. One strong point of contention that comes up occasionally during these discussions is the SSD:HDD cost parity claims many of the all-flash vendors are making. The position many of the all-flash vendors are taking is “my gear is incredibly fast, and costs the same $/GB as a disk based system!!” Sounds great, right? At first glance, I’d buy one too . . . until I figured out that the cost figures just don’t add up. The cost per GB comparison made by the all-flash vendors is being made against relatively low capacity 15,000RPM drives. If you want to see how the math adds up, I’ll give credit where credit is due. Scott Dietzen, CEO over at Pure Storage, lays out the numbers very well in his post titled “More Bang for Your Storage Buck“.
The bottom line street price he quotes in section 3 of his post says that Pure’s arrays are costing customers between $5-10 per GB. It is hard to know exactly what data reduction rates Steve used to get that range, but it we agree with Steve’s VP of Products, Matt Kixmoeller (another really nice guy) that, worst case, an Oracle database will yield between 4:1 reduction ratios and best case, a virtual server platform will yield up to a 17:1 reduction ratio, that puts the RAW $/GB for an all-flash array at anywhere between $40 and $85 per GB. That’s a pricey piece of storage. Now, for the sliver of data that needs the absolute best performance possible, perhaps that is OK – I am sure there are a few customers who would be elated with those economics. Conversely, the last time I looked at what I like to call “IO density” (IOPS/MB/sec) of your average enterprise datacenter, only 2-4% of the total capacity of a database or virtual server platform is begging for the blistering fastest performance possible. What happens to the rest of the data? What happens if that 2-4% is transient? These questions make a dedicated all-flash system a tricky solution to manage, right?
That’s why we believe a hybrid approach is best. All of our IO goes through high speed DRAM and SSD – up to 200,000 IOPS at about1ms latency fast – good stuff. The exciting part is, with a modest 3.5:1 data reduction ratio, our customers are deploying storage at $1/GB. Getting back to the analyst briefings I’ve been on, the consistent feedback I get when I explain our “200,000 IOPS at a Dollar a GB” is “WHOA! That’s awesome!”. That’s what I think too. Flash and SSD are disrupting the storage and virtualization market in a big way. I love being part of it. Being able to help customers enjoy the benefits of these huge changes in technology every day without having to pay $5-10 per GB makes me say “WHOA! That’s Awesome!” too.
Disclaimer: The IO Density Charts are re-crafted from publicly available data