One area that raises some questions from customers was around pricing. Put very simply, what is the secret sauce that allows Pure Storage to drive the cost of flash down to below the cost of performance disk, or <$5/GB usable? And what is the right metric to evaluate cost on a flash array? I’ll start by explaining the pricing, then I’ll explain the innovative technology that makes our low cost possible.
Straightforward $/GB usable pricing
In the new world of the modern storage array, differences in performance, utilization, HA/RAID overhead, deduplication and compression benefits all have a major impact on how much actual data can be stored on an array. This means that $/GB usable is quickly becoming the right metric to evaluate storage, not the legacy $/GB raw. This is discussed on our website, and is nicely articulated by Stephen Foskett.
So when Pure Storage calculates $/GB usable, we make some background assumptions:
All Pure Storage pricing examples are provided as HA, as this is a hard requirement for enterprise deployments
Pure Storage HA doesn’t rely on “mirroring” two arrays, rather it is true active/active clustered controller HA (add HA by adding a controller, not doubling the array or all the flash)
HA, RAID, metadata, and flash management have a combined overhead of 22% in the Pure Storage FlashArray and are included in our usable price (our RAID-3D is better than dual-parity, and can’t be set or adjusted, so this doesn’t vary from 22%)
All the prices quoted here are preliminary, and will be finalized with our GA release
Pure Storage pricing is, in round numbers (varies slightly by model and configuration):
$/GB raw for a HA FlashArray is $20 / GB
$/GB with HA and RAID is $25 / GB
$/GB usable (including HA, RAID, and 5-to-1 data reduction) is $5.00 / GB
$/GB usable (including HA, RAID, and 10-to-1 data reduction) is $2.50 / GB
Note that our Beta customers have commonly seen 5-10x data reduction in database workloads, and 10-20x data reduction in virtualization workloads, all without sacrificing performance. You can test your own potential data reduction on your data sets with our PRE tool.
Key questions to ask any vendor about pricing
Whether you are buying disk or flash for performance storage, ask these questions:
Are your $/GB prices raw or usable? How much data can I store on a XXTB raw array?
Do your $/GB prices include HA, RAID? If not, how much overhead is there to enable these? Do your datasheet performance numbers assume a certain level of RAID?
To enable HA, do I have to double my price by mirroring two independent devices?
If you offer data reduction, is it “always on”?
If you enable data reduction, does it impact performance? Can you achieve your datasheet IOPS/latency claims with data reduction on? Can you show examples of reduction rates on real customer data?
At a given performance goal, can I actually use all that capacity? For example, if I need 100K IOPS over 10TB of storage, do I actually need much more than 10TB of storage just to meet my performance goals?
Pure Storage technology innovations that drive down the cost of flash
OK, now that the prices are clear, you are probably wondering “how the heck did they do that?” The secret economic sauce at Pure Storage centers around three key innovations:
Purpose-built architecture for flash. We’ve designed our array from the ground-up for flash memory, and our entire architecture is optimized for flash, and in particular making flash enterprise-reliable and consistently performant. This allows us to reduce the cost considerably compared to legacy disk arrays which simply retrofit SSDs.
High-performance inline data reduction. We’ve build data reduction technology (compression and deduplication) right into the core of the FlashArray, so well-integrated in fact that it can’t be turned-off. It’s inline, it’s global, it’s high-performance (multi-100K IOPS at sub-millisecond latency), and it is designed to operate in the highest-end performance environments. This data reduction allows you to reduce your data between 5-20x.
Enterprise-level HA and resiliency. We’ve designed HA and resiliency from the ground-up for flash as well, as flash fails differently than disk. That means true active/active HA with clustered controllers, and a new form of RAID, RAID-3D which has three levels of independent parity optimized to the different failure cases of flash. Our HA and RAID implementation add a total overhead of 22%, again key to economics, when competitive solutions essentially forcing you to buy two arrays and mirror them for HA, doubling cost even before RAID is added.
These three innovations all point directly at how we achieve the price-point of offering solid state storage at below the cost of disk.
And let’s not lose sight of the big picture
Finally, I’d close by saying let’s not lose sight of the big picture here. Whether a vendor’s flash price is $5, $10, or $20/GB, all these prices represent a GREAT deal compared to inefficient spinning disk. Spinning performance disk today is $3-5/GB raw, and ranges from $5-20/GB usable after you add RAID, HA, required software, and all the spindles you need for performance. I hope that anyone interested in flash looks at Pure Storage Pricing, but more broadly, I think that the time is right for anyone who is looking at disk to look at flash. At these prices, why buy disk?