We just interviewed a candidate from a fellow upstart storage company. This gentlemen took a logo’ed coffee mug on every sales call. He told the customer to just display the mug prominently when meeting with their incumbent storage vendors to get a substantially better deal. With Pure facing a competitive brouhaha with EMC over all flash storage arrays (our thoughts on the XtremIO acquisition here), we quite liked this idea.
But a coffee mug is only the first step. We recently proposed three litmus tests that will help decide winners and losers in the all-flash array category. It turns out that you can exploit these same litmus tests to negotiate substantial savings in the procurement of all-flash storage (as well as flash tiers within hybrid storage). Here’s how:
- Ask your vendor if enterprise high availability (HA) including RAID (litmus test #1), deduplication and compression (litmus test #2), and zero overhead snapshots (litmust test #3) are on their roadmap. HA with RAID-6 (rather than mirroring), data reduction, and zero overhead snapshots are no brainers for flash—they reduce the flash media required to run typical workloads by 5X or more (the most expensive component of all-flash storage is the flash), and they eliminate unnecessary writes (which are expensive in both flash wear and performance).
- Assuming your vendor says yes, then share your workload’s data reduction ratio—that is, the estimate of how much raw flash is required to store your dataset provided the above optimizations are implemented. You can find a free software tool introduced here to make this easy. The tool yields the ratio of usable GBs of your data you can store per raw GB of flash, with parity and meta-data overhead included in the calculation. (Note: It is important for you to measure a reflective dataset—one that includes roughly as many virtual machines or database instances that you would like to maintain on the storage going forward. Larger datasets tend to have greater reduction ratios, and hence greater negotiating leverage for you.)
- Then ask your vendor for the discount necessary to keep you whole for the additional flash you have to buy because they have not yet passed the three litmus tests. After all, why should you pay extra in the interim until your vendor gets around to doing their HA, dedupe and snapshot homework? For Pure’s target virtualization and database workloads, the above could give you the leverage to negotiate 33%, 50%, and even 80% discounts. After all, the last thing your current storage vendors want is a new competitor in your data center, but it is likely a boon for you.
These simple negotiating techniques are reflective of a broader change coming in the performance storage market. Today, when you buy performance storage (whether that be flash or 15K spindles), you typically quantify your purchase in dollars per GB raw—that is, the total purchase price divided by the total GBs of storage media. On the other hand, when you buy a disk-based backup appliance, you are likely to quantify your purchase in dollars per GB usable—the total price paid divided by the amount of user data you can store on the device taking advantage of deduplication and compression.
The same thing is now starting to happen with performance storage. Suppose you’re buying PCIe flash cards for your servers. Further suppose you intend to protect your data via mirroring. Then the PCIe flash card that costs you $10/GB raw is actually $20/GB usable. Storage arrays are no different. There are many performance workloads that still demand RAID-10 with mechanical disk, because the more space efficient RAID-5/6 alternatives are just not fast enough. In that case, $2.50/GB raw becomes $5/GB usable. Short stroking for performance contributes as well: If your hard drives end up only 2/3 full, then your $/GB usable is 50% higher. Ditto thin provisioning. As a result, it is not unusual to end up with a $/GB usable that is 3-5X higher than your $/GB raw price. Of course, your vendor would rather price what you’re buying at $2.50/GB raw than at $7.50-12.50/GB usable.
Purpose-built all-flash arrays provide thin provisioning and never need to be short-stroked, and yet there can still be substantial gaps between $/GB raw and usable. In fact, there are all-flash arrays that use simple mirroring rather than RAID 5/6, even though flash memory tends to have sufficient I/O capacity that there is effectively no performance penalty for RAID 5/6 schemes. This leaves a customer in the silly position of having to employ mirroring (100% space overhead) rather than RAID 6 (~25% overhead), even though the latter provides stronger (dual parity) protection and may well be faster (reduced writes). In our view, mirroring within a flash array simply doesn’t make sense: e.g., $15/GB raw becoming $30/GB usable. Far better, then, for you to insist on paying for the value that accrues to you—the $/GB usable, rather than risking the vendor hiding the true cost of their product for your workloads.
Just as within the backup market, deduplication will be the straw the breaks the $/GB raw metric’s back. Flash memory is a random access device like DRAM rather than a sequential access one. As a result, flash admits new high-performance, high-resolution algorithms for deduplication that do not slow down performance. In fact, dedupe on flash can increase performance by not writing the same data over and over the way it is done on performance disk (writes are expensive on flash). With deduplication and compression delivering substantial savings, our claim is that in the future you won’t buy flash storage without it. (Note: These data reduction techniques appear to be generally incompatible with mechanical hard drives running performance workloads—Deduplication is so random I/O intensive that traditional disk cannot hope to accomplish it with submillisecond latencies—see Not Your Momma’s Deduplication and Why Flash Changes Everything.)
The stunning result we have seen with Pure’s customers is that their average $/GB usable with all-flash is typically less than their $/GB usable with hard drives! We have numerous endusers who are paying $2-3 GB/raw for performance disk, but find their $/GB usable is in the $6-15 range (counting incremental software licensing fees). With the Pure Storage FlashArray, we have flipped these ratios: the FlashArray is relatively expensive compared to disk in terms of $/GB raw, but for our target virtualization and database workloads, we are seeing $5-10/GB usable, less than the comparable cost for performance disk (without a flash cache/tier)! Moreover, the above analysis doesn’t count the savings from flash’s greater power efficiency. Nor does it count any savings from server consolidation, lower DRAM requirements, or reduced enterprise software licensing fees (as a result of server consolidation). Best case: When you factor these in, you may well find, as some Pure customers have, that you can deliver a 10X performance and save enough money that your flash storage pays for itself! Worst case: You gain additional negotiating leverage with your favorite storage vendor.
With all-flash storage faster, more space & power efficient, simpler and cheaper, why buy disk?