Tomorrow EMC (NYSE: EMC) is expected to announce the general availability of the XtremIO all-flash array. In doing so, the storage market leader is arguably demoting mechanical disk to capacity rather than performance workloads. To their credit, EMC has managed through innovator’s dilemmas better than most. Their top tech evangelist, Chad Sakac, recently said in his blog:
Flash is changing everything about storage … Sounds like hyperbole? It’s not. Sometimes people don’t “get it”. Flash is like someone coming along and saying “I have a new car on the market. It’s pretty awesome. It fits in your pocket, but can still get you around. It’s gets 1000 mpg. Oh, and it can go 10,000 mph.” That would be a pretty radical inflection point, no?
It’s no surprise that the all-flash array category is the fastest growing in storage history. Both Pure Storage and XtremIO use data reduction to bring the cost of all-flash arrays in line with those of disk-centric arrays—that is, the same dollars buy you technology that’s at least 10X faster, 10X more space & power efficient, 10X more reliable, and 10X simpler than the mechanical alternatives.
Hybrid arrays (of flash and disk) just can’t keep up. With hybrids, the latency of the slower media dominates: doing the math, if flash is about 100X faster than disk and you miss the flash cache just 5% of the time (95% cache hit ratio), your performance drops over five-fold! So hybrids still perform like disk, but generally cost more than the all-flash alternatives from Pure or XtremIO (in terms of $/GB usable for systems like VNX or VMAX). For virtualized and database applications, all-flash arrays are a no brainer, and that’s the majority of the $15B annual spend on performance (oxymoron) disk storage. As the storage market leader, EMC’s endorsement of the “all flash at the price of disk” value proposition is only going to accelerate the inevitable transition from mechanical to solid-state.
The GA of XtremIO is also going to bifurcate the all-flash market between the software have’s and the have not’s. It is no accident that XtremIO and Pure Storage look very similar. The winning recipe for all-flash storage is set:
- High Availability (HA) combined with dual-parity or greater RAID protection – Mirroring is nonsensical on flash because it’s 100% overhead for the equivalent of only single parity protection and no appreciable performance gain. With dedupe, you really need greater than single parity protection, since without it, overlapping failures in the media storing highly deduped segments would be catastrophic.
- Non-Disruptive Upgrade (NDU) of controller software, component firmware, and hardware (both for replacement & expansion).
- Inline, submillisecond deduplication – With flash, there is no random I/O penalty and writes slowly wear the underlying media, so it just doesn’t make sense to write the same data over and over again the way it is done on disk. Data reduction expands storage capacity, increases performance and the extends the life of the flash.
- No overhead snapshots & clones – Most vendors are viewing dedupe as an after-market add on. When dedupe is instead built into the foundation, as it is within Pure and XtremIO, you can deliver near zero-overhead (both in space and performance) copying of volumes, as this entails only an update to the meta-data. (And similar benefits accrue to native replication, which is coming soon from both Pure and EMC XtremIO.)
This all seems obvious in retrospect, and yet Pure and XtremIO are the only all-flash arrays offering this feature set today, and I submit that our mutual competitors are still a long way from delivering the above recipe (despite the fact that these items are now included in most vendor’s roadmaps). The last place you want to be as an early-stage company is looking up at EMC because you are behind on critical features, but that is arguably where the rest of the all-flash field wakes up tomorrow, with Pure Storage the notable exception.
At this point you might question my sanity: Why would the CEO of Pure Storage be giving our principal competitor EMC so much credit? As we have remarked before, Pure welcomes competition. Competition drives innovation and customer value. Competition makes us a better company, better at meeting the needs of our customers and our partners. Competition also fuels market growth. We are excited that EMC’s principal flash storage offering in XtremIO is finally entering the market, as we expect both Pure and XtremIO to benefit hugely as customers redirect their purchases from mechanical to all-flash storage.
At the same time, we remain very confident that Pure will continue our winning streak against EMC XtremIO. Here’s our top ten list why:
(10) Hardening – Pure has been running production workloads for 3 years, whereas XtremIO is just now being readied for production deployment. HA and NDU take many months of hardening on test and customer workloads. Pure has been competing with XtremIO within numerous customer data centers, and to our knowledge EMC has yet to demonstrate HA and NDU. Before deploying XtremIO or Pure, insist on testing both HA (pulling drives, cables, power cords) and NDU. It’s the only way to be sure the technology is ready for your business critical applications.
(9) Persistent metadata – Like XtremIO, Pure caches its metadata (index of what is stored where) in DRAM. Unlike our understanding of how XtremIO works today, we also synchronously persist I/O and metadata to nonvolatile flash before acknowledging a write (and do so redundantly). Anyone that’s ever misconfigured a UPS or had one fail when it was needed, will be inclined to think that stronger metadata and write protection is required than what XtremIO is shipping currently.
(8) Lossless deduplication – Pure read verifies all duplicates, as has been the standard practice in primary storage (NetApp) and WAN optimization (Riverbed). Relying solely on hashes without read verifying dedupes would lead to data corruption in the event of a hash collision. Although the chances of a collision are very rare, such shortcuts have historically not been acceptable for the primary copy of user data.
(7) Compression – Across the 100s of Pure arrays deployed around the world in mixed workloads (not just VDI deduplication hero numbers), Pure averages 6:1 data reduction. And that 6:1 does not count savings from thin provisioning (zero removal). About ½ of that savings comes from deduplication, while the other ½ is derived from compression. Via the combination of both techniques, Pure customers get about 2½ times greater data reduction, and the confidence to deploy flash for workloads that benefit from either dedupe or compression or the combination of both together.
(6) Flexible scale (up & out) design – Pure’s architecture was designed to support independent upgrades for both performance and capacity, as it is unusual for increases in each to be needed simultaneously. Pure supports adding a shelf to increase capacity at the lowest cost possible, and supports annual non-disruptive controller upgrades to increase performance (with Moore’s Law). We’ve also designed scale-out into our architecture for tomorrow’s million+ IOPS workloads. XtremIO, on the other hand, offers only scale-out, requiring an expensive controller addition for every capacity expansion, leading to higher cost, lower density, and overspending for unused IOPS.
(5) Zero performance impact NDU – It’s ironic that most storage arrays slow down substantially while they are being upgraded—How is it a non-disruptive upgrade if performance drops 50% or more? Pure provides NDU without performance degradation by reserving enough capacity in our active/active design to accommodate seamless upgrades. Before buying any NDU array, you should perform an upgrade under load to see what happens.
(4) Variable block size – XtremIO uses a fixed alignment 4K I/O size, which generally does well on artificial benchmarks, but loses ground on real-world virtualization and database workloads, where larger I/Os are typical. The average across Pure’s customer base is between 32K and 64K sized I/Os, which represent 8 and 16 independent I/Os respectively for XtremIO. (Pure also deduplicates at variable chunk size, which allows larger and smaller dupes to be captured more efficiently—such as those within VMs, as well as the capability to find misaligned dupes (from VMs or applications that aren’t 4K aligned) that XtremIO would miss with its 4K-only approach.)
Trust and Investment Protection
(3) Security – Pure supports full data at rest encryption with zero performance overhead or key management, as well as Active Directory integration and role-based access control today.
(2) Channel friendly strategy affords greater market access for Pure – EMC is roughly a 1/3 market share player, and they do about 1/3 of their business through the channel. That means about 1/9 of all storage systems are sold to endusers by EMC channel partners, leaving 8/9 of the market available to Pure and our channel partners (we are >90% channel today).
(1) Leverage the Innovator’s Dilemma – First movers tend to win in tech—that is, David usually beats Goliath, and does so because software differentiation has proven resilient. Pure crafted the original recipe for all-flash replacing performance (Tier 1) disk. Remember Data Domain vs. Avamar? EMC bought Avamar early while Data Domain elected to stay independent as Pure has chosen to do. Which technology was the better long-term bet?
More importantly, XtremIO is a nascent product that will ultimately cannibalize EMC’s flagship VMAX and VNX product lines. This is a difficult minefield for EMC to navigate. If you want to be sure to get access to the latest and greatest technology from EMC, perhaps your best bet is to have a Pure Storage coffee mug on your desk (maybe an extra 5 points of discount?) or a Pure Storage array in your data center (an extra 15?) ;-).
Conclusion. For EMC customers and partners, if you want to see what XtremIO is likely to look like tomorrow, check out Pure Storage today.
If there’s any question that EMC XtremIO is targeting Pure Storage, I will share that unbeknownst to us, about a year ago EMC procured a Pure Storage FlashArray from a channel partner and did their own internal testing. Pure has never undertaken anything of this sort. Our understanding is that the Pure Array in question was taken to an engineering facility and that some of the XtremIO team were able to get hands-on access to Pure’s technology. We believe that such practices risk intellectual property contamination, both perceived and actual, of any vendor who allows their technical staff direct access to a competitor’s technology. Customers tell us there are still head-to-head performance testing claims circulating from EMC, perhaps based on the information gathered from this testing (some of which we consider to be our confidential and trade secret information). This, despite the fact that the Pure array in question, was of a previous generation and any such comparisons would be apples to oranges. Pure eventually did get the box returned after haggling, although it was damaged somewhere along its journey.
The lesson herein is to not rely on marketing comparisons, such as my own arguments above. Rather you should use such insights from competing vendors to inform your own test plans. At Pure, we welcome head-to-head hands-on compares between Pure Storage and XtremIO, or any storage array for that matter. Worst case, the return on your time invested will be a better deal on your next storage purchase via cranking up the competition. Best case, you’ll get a better, more mature product as well as something that’s 10X better across the board vis-a-vis the disk-centric alternatives. With Pure’s unique Love Your Storage™ money-back guarantee, the risk is actually in not taking a closer look at the future of performance storage before you throw good money after bad on the mechanical disk legacy.
*The performance and technical features attributed to XtremIO set forth in this blog represent our understanding of publicly available information as of press time.