IOV tech could give Violin storage platform status
Flash array startup could pull off a NetApp
Blocks and Files Violin Memory might well become the next platform storage company. But to do that it needs three product technologies and it only has one at the moment: its shared all-flash Memory array.
The context to this is that primary data storage for latency-sensitive and IOPS-sensitive applications is moving away from spinning disk to flash memory. Disk drive arrays take too long to find data and too long to deliver it across networks. Their IOPS numbers are too low unless you throw lots of short-stroked spindles at the problem, and then the cost rapidly becomes huge, with $/IOPS going way above that of a flash array.
Shared flash arrays that use multi-level cell flash and deduplication arrive at a $/GB figure that may be five times the $/GB of a disk drive array but they can deliver tens of thousands of IOPS and an acceptable working life with endurance-sapping writes minimised through wear-levelling, RAM buffering of incoming writes, and sophisticated garbage collection. The deduplication processing increases the latency of the flash array but decreases the cost/GB and so is an optimisation.
Now let's return to Violin Memory and its coming TLC, 3-bit multi-level cell memory (MLC), which is now 50 per cent more capacious than 2-bit MLC, courtesy of Toshiba. This, I understand, will make Violin's Memory array only twice as expensive in $/GB terms than the $/GB of 15,000rpm disk drive storage. That's good news, not enough to propel Violin to platform storage status, but a boost nonetheless.
There's more though.
Server flash card
I understand Violin has developed a server flash card product.
This is a PCIe-connected solid state storage module that plugs directly into a server and provides data access with no network latency at all. It means Violin will directly compete with Fusion-io and its PCIe-connected ioDrive flash hardware, as well as every other server flash vendor. If we imagine Violin possibly providing a range of server flash cache cards, from SLC performance-optimised to TLC capacity-optimised, then that would be a neat product set, but still not enough in my opinion to enable Violin to become a platform storage play.
There's more though.
Violin would be competing particularly with EMC which has been the most vociferous and active storage array vendor concerning server flash. It has its VFCache server flash cache product and also its coming Project Thunder shared all-flash array, and it's also buying XtremeIO for its performance-optimised, deduplicated, shared and scale-out all flash array. This seems odd as having a very fast networked all-flash array means you still suffer from network latency. You really want the fast stuff as close to the server as possible.
This is Basile's big play. Can he pull it off?
Server I/O virtualisation
If the fast flash stuff has to be as close to the server as possible and yet still be sharable, meaning networked, then that network has to be PCIe-speed and PCIe-connectable. So far Virtensys, the PCIe fabric extension company, has failed to light a fire and had to be bought by Micron. Aprius, a server I/O virtualisation (IOV) start-up, crashed and burned. Xsigo, a PCIe-over-InfiniBand IOV startup is not making huge waves although it is full of excitement.
We know Violin recruited an Aprius co-founder, Peter Kirkpatrick. Word has reached me that Violin has IOV technology. It can, it seems, connect its shared memory arrays to servers at PCIe speed and make their contents part of the server's memory space. This is huge.
Let me say that again; this is huge.
SAP recently invested in Violin with the obvious prospect being Violin's flash products being used to store HANA, SAP's in-memory database. There is no point in putting an in-memory database out in an array across a network link. Think about it; the Network link adds the latency that being "in memory" is supposed to take away.
Why did SAP invest in Violin? What did it see in Violin's future that would tempt it to invest $10m or more in Violin?
Another question. Why isn't EMC buying Violin Memory? It surely has enough cash to do that if it wanted? Could it be that EMC has misread the market and thinks the shared flash array has to be very high-performance, which is true, but doesn't need IOV to hook it up to the servers? Unlikely isn't it; that would mean EMC being in denial of something that seems obvious. If that is not the case then EMC could have access to IOV technology too.
My thinking is that EMC isn't buying Violin because Don Basile, Violin's CEO, and Violin's board want too much money, and here I'm thinking of several billion dollars. They want to build a multi-billion dollar storage platform company, a feat that was last pulled off by NetApp.
There is more pertinent context here.
Flash arrays will eat primary data storage lunch
Every storage array vendor now understands and believes that primary data storage for performance-bound applications is moving to flash. Using flash to best effect means that current disk drive array controller technology has to be rebuilt from the ground up for flash.
Neither VMAX, VNX, V7000, FAS, whatever, can do anywhere near as good a controller job with flash as can controllers architected from the ground up for flash. The Whiptails, Nimbus, Solid Fires, NexGens, Pure Storage, Greenbytes, XtremeIO and Violin Memories of this world will eat every primary data drive array controller's lunch and still be hungry.
All the primary data disk drive arrays are going to become gigantic data tubs, full of SAS or SATA drives feeding and backing up primary data storing flash arrays. This is already beginning to happen: Violin Memory, I hear, is winning deals from EMC over primary data storage for performance-bound apps. It's a movement that will accelerate over the next five years. Every mainstream storage array vendor has to get networked flash array and server flash technology or fade away into insignificance: every single one, without exception.
That's why EMC needs a stake in the all-flash, shared array party, and why there is going to be a feeding frenzy as the other shared storage disk array vendors realise they need to act as well. It will time for them to do it, and then more time for them to assimilate the technology and integrate it with what they have. While that is going on Violin will be using is recent exec-level recruits to get itself big data, VDI, analytics and other software technology it needs to become the flash storage platform it wants to be.
Violin Memory CEO Don Basile
This is Basile's big play. Can he pull it off?
Violin as a platform
Who is going to stop him? Fusion-io doesn't have its own shared array technology, with start-up partner NexGen doing that job.
None of the mainstream disk drive array vendors can take on Violin directly. They don't have the technology and have primary data-storing disk drive array product revenue streams to protect. And, apart from Nimbus and Whiptail, none of the other shared, all-flash array startups are actually selling shipping product. I'm told Whiptail, not having deduplication, is not in the same league as Violin.
Violin could pull off a NetApp, if it has the technology building blocks I'm told it has and if it can execute right. These are real big "ifs" and I could be being fed stuff to big up Violin's upcoming IPO. There is always that risk. The S1 for this, by the way, has been filed, and the IPO could take place in the next month or few.
Our storage world could be turned upside down by flash and IOV. No, let's go further; it is going to be turned upside down by flash and IOV. What exciting storage times we live in. ®
Sponsored: Benefits from the lessons learned in HPC