Ahh it’s that magical time of year again, when aspiring flash memory companies hawk their best stories and bring out their wares, some real and some imagined (i.e. roadmaps) for the review of the flash industry analysts and press world-round. We’re going to be enjoying this round of the media frenzy as spectators from the sidelines (since we announced and shipped our third-generation FA-400 FlashArray last quarter, and have already been shipping GA to production customers in volume), but that won’t stop us from sharing some observations about the latest news. Here are four key observations on the state of the all-flash array market, and if you’d like to hear more live, please stop by and see Scott Dietzen, Pure Storage’s CEO, give a live keynote at the FMS on Thursday at 2PM.
1: It’s still all about the software.
If there’s a mantra we have at Pure, it’s that we are determined to be the most talented software team in flash storage – bar none. Broadly-speaking, you will see in the market two types of flash storage companies: 1] “Flash Appliances” – hardware-centric innovators who deliver custom and often dense hardware designs, but for whom software is always an afterthought (their products resembling really big SSDs, or “JBOFs”), and 2] “Flash Arrays” – software-centric innovators, who invest their IP in innovative software around flash to deliver the full stack of enterprise storage array features, leveraging the advances in industry-standard hardware (both SSDs and CPUs). Pure is quite decidedly in camp #2, our belief being that while it is possible to carve-out a short-term edge through deep hardware innovation, ultimately in most markets hardware commoditizes, and what you have left in the end is software differentiation. In this market, although flash hardware is fantastic, it is the software that will help overcome the real customer barriers:
- cost (data reduction via deduplication, compression, and thin provisioning)
- resiliency (high availability – HA, non-disruptive upgrades – NDU, snapshots, backups, replication)
- longevity (write coalescing, wear leveling, and write avoidance through data reduction)
- scale for consolidation (up/out scale and expandability online, piecemeal growth, multi-array management)
- adoptability and integration (application integration, VM-level integration like VAAI/ODX, protocol support)
Our strategy is simple: focus on leading the software race around flash, and leverage and drive hardware innovation with our partners in the flash/SSD and CPU communities in tandem.
2: Data reduction technologies – where are the results?
When Pure started in 2009 one of our founding realizations was that data reduction software (deduplication, compression, pattern removal, thin provisioning) was really hard stuff, and a key technology for flash (lowering cost + delivering write avoidance). Our belief was that neither disk nor flash vendors who didn’t design their products for data reduction from scratch simply wouldn’t be able to effectively retrofit it to their products, delivering on the six key design criteria of effective data reduction software:
- inline (to avoid writes),
- global across the entire array,
- without performance penalty,
- both dedupe and compression to serve databases and VMs/VDI,
- at ultra-fine chunk sizes to auto-align with all applications
- without trusting probabilistic hashes, which have a chance of collision (i.e. data corruption).
So here we are four years later, and it seems like everyone in the flash space has a data reduction story, and are positioning data reduction technologies as a “must have me-too feature”. Well, when you check the report cards of the industry, we see spotty implementations across the board: some products with dedupe but no compression, some with compression but no dedupe, some that suffer massive performance hits when data reduction are enabled, some that are on their 2nd or 3rd try at OEMing and retrofitting dedupe technology, and NO other vendors who have shared any public information about their delivered data reduction rates other than Pure Storage.
I find the last point particularly interesting…in May Pure Storage launched our “Dedupe Ticker” (just check our home page at www.purestorage.com) – displaying real live data reduction results, averaged across EVERY Pure Storage customer with phone home enabled (the vast majority). As you can see our data reduction is delivering >6-to-1 savings (>12-to-1 including thin provisioning savings)…and no other vendor has followed-suit in sharing this level of transparency. These results are particularly compelling, as they are all real-world production workloads, and across a wide variety of use cases from databases to VMs to VDI.
3: What’s more important, ultra-density or resiliency?
100s of TBs in 1U. Neat trick, but is ultra-density really the most important design point for the mainstream market? Would most customers bet better served with vendors optimizing first for ultra-density or resiliency?
When we started working on the Pure Storage FlashArray, we looked at a variety of hardware designs and form factors. Broadly speaking – there were two options: ultra-dense appliances (use DIMM-like top loading designs to achieve higher density) or more traditional carrier-based shelves (provide hot-swap ability for service and expandability online).
Ultimately we decided (as did all of our mainstream competitors) that Reliability, Expandability, and Serviceability (RES) were more important than ultra-density for the markets we are targeting (enterprise and SaaS/IaaS). After all, we are typically replacing literally multiple refrigerators of spinning disks with something the size of a microwave….is it really necessary to make it VCR or pizza-box sized, if that means sacrificing RES? Some questions to ask yourself, and to ask your flash vendors:
- Do you care about real high availability, with independent controllers and components, or do you trust your redundant components all in one box connected by a single passive or active motherboard? Have all the failure paths been tested? As a consumer, how can you test all the failure paths to gain comfort yourself? Do you just have to take the vendor’s word?
- Do you care about non-disruptive upgrades, or do you want to take downtime for every application whenever the array needs to be upgraded? Does the architecture even support non-disruptive upgrades? NDU has been standard-issue in storage for over a decade.
- Do you care about expandability? Do you want to fully-populate your flash array on day one, of would you like to start small and expand as you grow without downtime? Does your vendor’s architecture support hot-expansion?
- If a flash module fails, how do you plan to replace it? Would you really slide a running flash array out of its rack, open the top cover, and “play operation” to remove a failed module while your key business applications are running? If so, please give this a shot during a PoC, and let us know how comfortable you feel about it!
Net net: density is important to the extent that it can be used to drive significant cost out of the architecture, and Pure believes in continually pushing density for this benefit, but not at the expense of sacrificing resiliency, expandability, and serviceability. The sweet spot is to maximize density without sacrificing RES.
4: What happened to the storage establishment?
And to end on a lighter note, 2012 was the year of “big storage” waking-up from their slumber to join the all-flash storage revolution….we saw EMC acquire (but not ship) XtremIO, NetApp announce their plans to build FlashRay, IBM acquire TMS, Hitachi build their own flash controller, 3Par announce a flash storage retrofit disk array, and Dell threaten to do something with flash a few times….but alas, none of them are showing their wares at the show. Maybe next year?