Pure Storage Tackles Storage Shenanigans
Pure Storage, a veteran sponsor to Gestalt IT’s series of Tech Field Days, presented at Virtualization Field Day 2 out of its headquarters in Silicon Valley. The company is creeping up on a GA release of their all-flash storage FlashArray products: the FA-310 and FA-320. I last wrote about them from Tech Field Day 8, and aside from checking off the list of road map items as more customers and partners have put their workloads on beta releases of the product, not too much has changed. In this post, I’ll give some props to the very clever use of the Foskett VMware Integration Scale, along with some discussion I had with Matt Kixmoeller, VP Product Marketing, on how Pure Storage is different from other players in this space. By the way, Matt, your twitter photo is a little creepy with the logo on your forehead. Just sayin.
Heard of the Foskett VMware Integration Scale?
I will admit that I had not (Sorry Stephen). But I have now! And I agree with the arc. Basically, it preaches getting on the HCL (Hardware Compatibility List), making a plugin, and then going hog wild with supporting VAAI (vStorage APIs for Array Integration) primivites.
I’ve always felt the HCL was sort of a money grab. If you want to be considered “Enterprise Grade”, you better pay to join the list. If not, no one will take you serious. It’s just a right of passage for anyone who wants to do business with any vendor, not just VMware. Yeah, it can be a safety net to help ensure you’re buying something that’s “compatible”, but at the end of the day, what does that really mean? How could this matter for storage – it uses a standard protocol (fiber channel) to communicate to servers, can’t get much more compatible than that. Maybe I’m missing something?
Pure Storage has commented on their progress underneath each objective in orange.
As for plugins, if you run NFS – you must have this! There’s no way to do VAAI without a plugin. In Pure Storage’s case, it’s not that big of a deal if they deploy a plugin as they aren’t shipping with NFS, but they did comment that a plugin is in the works later in the game.
Personal Rant: A plugin that simply loads a web page is not a plugin. Try again. (Not directed to Pure Storage)
Lastly, the support of VAAI primitives. Today, the product already supports:
- Atomic Test & Set
- Block Zeroing
By Q2 GA they plan to also support:
- Thin Provisioning Stun
- Quota Exceeded Behavior
- TP LUN Reporting
Extended Copy will be supported sometime after GA.
It’s All About The Benjamins
One slide I chuckled at was a Forrester quote (For those playing at home, you can take a drink) stating “If cost were no object, you would put all your data on flash”. My response to this is a very deep and eloquent “Duh”.
Pure Storage battles cost in a few ways. The primary attack is through deeply integrated, in-line deduplication. It’s on by default, and John Hayes, Founder and Chief Architect, was kind of boggled to think of why you, the customer, would ever want to turn it off. I agree. Based on 35+ customer deployments, Pure Storage is stating that, on average, you should see about 8.46:1 data reduction for “virtualization” workloads and 4.85:1 with “database” workloads. I’m not quite sure what the mix of servers are in these scenarios, as a VDI deduplication number is always going to be stellar, whereas a customer running 48 flavors of application tiers will be a bit low. Still, anything above 2:1 seems pretty good, so even a deduplication pessimist can hope for that or better.
The other tactic is using commodity MLC flash and baking in all the intelligence into how they handle the data. This helps drive down cost, while at the same time enables Pure Storage to “carefully manage the MLC flash, ensuring long life and consistent performance“. The fact of the matter is that demand reduces cost, and it appears that MLC flash, through consumer demand, is winning that battle over SLC.
I’m a sucker for the word shenanigans, especially after watching Super Troopers so long ago. Pure Storage is targeting a number of those shenanigans that have come about because of storage being an issue. Two resonated with me rather well, and I feel the winds of change coming about with other companies also embracing these ideas.
Super Troopers. Best … movie … ever.
Shenanigan #1: Many Small LUNs & Datastores
The idea of having larger sized datastores (in this case, LUNs) with mixed workloads seems to be gaining in popularity, or at least getting some buzz time. When tiering is no longer an issue (courtesy of having everything on SSD), nor is datastore size an issue (courtesy of ESXi 5), provisioning storage becomes a bit less complex. Will your storage admin like these changes? I’m guessing not. It will be an interesting paradigm shift.
We sure have come a long way since these days
Shenanigan #2: Over-Provisioning of Server DRAM
The other shenanigan was the thought of using server DRAM to act as storage, such as with a database or other transactional application. I’ve seen this first hand, where storage is just too pokey or provisioning enough spindles to cover performance wasn’t practical. Pure Storage hopes to put an end to this, and I’m all for it. Monster VMs of this type are more like Frankenstein VMs.
As with Tech Field Day 8, I really enjoyed the time spent at Pure Storage. These guys never fail to deliver a great presentation with a live demo and full disclosure. For another pair of eyes on the presentation, head on over to fellow VFD2 delegate Dwayne Lessner‘s post at IT Blood Pressure, which discusses some thoughts on VDI with Pure Storage.
Oh, also – I snuck down the hall to snap a photo with the neighbor’s … uh, purple guy blob thing.