FLASH: The Disruptening
Who will tame this mighty beast and ride it into town?
Comment We have had a flash of insight: all the storage array vendors are going to have to face up to all-flash arrays and do something about the technology – buy, partner or build. Denial is not a strategy.
Just as disk storage bifurcated into server/PC direct-attach storage (DAS) and shared networked storage (NAS and SAN), so too will flash come to be used in direct-attached and shared modes, either attached to a single host or shared between multiple hosts (servers).
The various storage array flash caching designs – such as accelerator boxes in front of the array, flash cache in the array controller, and solid state drive (SSD) in arrays as a tier zero for the hottest active data – will be eclipsed by a return to basics, to flash as DAS or flash as NAS/SAN. Why should it be any different from what has happened with disk drive-based storage?
Incoming players upset storage array applecart
Fusion-io is leading the flash DAS market with its PCIe-connected ioDrives and its virtual storage software layer to pool ioDrives together into a multi-terabyte flash pool. Fusion's eight to 10 followers are pushing the PCIe flash drive idea strongly and it is on the verge of becoming mainstream. There is nothing the storage array vendors can do about this, apart from trying to manage server flash caches themselves, but this may well be a self-limiting idea.
On the networked storage front Nimbus Data and Huawei-Symantec have introduced all-flash arrays and Violin Memory, flush with funds, is punting its flash memory array technology strongly. EMC has said it can produce all-flash VNX configurations. It seems abundantly clear that, on a performance level, flash arrays make hard disk drive arrays look like people-carrier vehicles trying to compete with Ferraris. Give up disk array guys. It's a no-brainer. You are going to lose.
As soon as flash arrays are affordable and reliable, they will become the natural repository for all primary data currently stored in the disk drive arrays. Nimbus says that day has come already and cites its eBay win against NetApp and 3PAR. Violin Memory is licking its lips in anticipation of 3-bit multi-level cell flash causing flash array prices to drop substantially. Oracle can do all-flash arrays now too.
Let's cede this case and say that all the storage array vendors – BlueArc, EMC, Dell, Fujitsu, HDS, HP, IBM, NetApp, Nexsan, Oracle and everybody else – face having their customers buy all-flash arrays and putting these in front of their existing storage arrays. These then become nearline stores, not primary data stores. The array vendors face having their strategic relationships with customers weakened; the flash arrays become strategic while the disk drive arrays are just spinning rust warehouses for less active data, the boring stuff.
What can the spinning disk storage array vendors do?
Existing array players' strike-back strategies
It's obvious, isn't it? They build, buy or partner to get the the technology. Here's a light-hearted look at what the individual vendors might do.
BlueArc basically sells hardware-accelerated NAS arrays, based on founder Geoff Barrall's inventiveness. It's debatable if the company now has the same raw creativity and ability to carry through a build-from-scratch flash array strategy. It will likely buy a flash array supplier and integrate its product under a BlueArc management and data protection software layer.
EMC, the arch purveyor of separate storage products for separate storage apps, can bring out all-flash VNX and VMAX configurations but could, and probably should, and probably will, buy a flash array startup. One of the choicest low-hanging fruit is Violin Memory, flush with funds from venture capitalists who are always interested in money-rich exits.
Violin's strongest competitor, and possibly the market leader, is TMS with its RamSan products, but TMS is a venture capital-free zone, with no board-level pressure for an exit. Newer flash array startups like SolidAccess and SolidFire are relatively unproven. Either TMS or Violin would give EMC, with its channel power, dominance of the flash array space at a stroke.
Dell: It's a toss-up between the new-found buying zeal exhibited by Dell or its traditional partnering stance. If we assume the company now believes it has to own significant technology then it will buy. The company tends not to win high-end bidding wars, witness 3PAR, so it may try to buy a second-string player, such as Nimbus, or one of the newer startups like SolidFire.
Japanese Fujitsu will move cautiously and relatively slowly, and will likely partner. It could, for example, do a deal with TMS.
HDS will partner with a flash array supplier in preference to buying one because buying in technology seems a difficult thing for this Japanese-owned firm to do. It has a track record in partnerships, with BlueArc the obvious example. It could, for example, also do a partnership deal with TMS and sell RamSans into its customer base.
HP will want to present its all-flash array as an integrated part of its converged infrastructure (CI). To do that it will have to own its own technology and will, we guess, buy a flash array startup. We think it could perhaps shoot for Violin or TMS, or go for an early stage startup whose technology could be more easily melded into the CI scheme.
We note that HP has its own post-flash storage technology in the shape of the Memristor. This we think is too far away from productisation for HP to base an all-solid state storage array strategy on it right now.
IBM: Big Blue could take a NetApp all-flash array if Sunnyvale's engineering teams react quickly enough. Otherwise IBM will probably buy in the technology – a strategy it has pursued many times before in the storage space.
NetApp could buy but it could also build. For example, it already has a detached FAS controller head in the shape of the V-series, whose role is to virtualise third-party arrays and bring them into the NetApp fold. We could imagine a V-Series with a huge stash of flash acting as an all-flash array in front of NetApp's FAS arrays and third-party arrays.
If NetApp does this, an all-flash V-Series would be manageable underneath the same NetApp management umbrella as all the FAS boxes and their software. NetApp would have to add data moving software to ship data between the flashy V-series and the existing FAS boxes but that could be done after the all-flash V-Series product is introduced ... if NetApp introduces it.
Nexsan could build is own speedy and affordable all-flash box, and call it something like the SAS or SATA FlashBeast.
What about Oracle? Been there, doing that already. Larry will build. He doesn't believe in OEM or reseller deals, and the Exadata, ExaLogic boxes show what Oracle can do.
A stray thought: an all-flash Pillar Axiom could be an interesting proposition, re-energising the brand and giving it a new lease of life, if Pillar engineering can act fast enough.
Here at Vulture Towers we're convinced that flash arrays will become mainstream. All the existing storage array vendors have to embrace the technology somehow. This is like the arrival of deduplication technology – and look at the buying frenzy that generated. Flash array technology is going to be disruptive, that's for sure. ®
Full stops are free.
but they wont necessarily prevent the rate of cascading domini/domino effects in motion, without of course a skirtless mega moderatrix mandate machine made merely to Allow/Ensure/Inform/Outperform/Unite and Inclusively Improve Investiture Immensely for any/all hearts and minds in L.O.V.E.
An amnesty and a silence will be required for the truth to be heard by the herd.
Forgiveness is a healing virtue, once the harm is measured, managed and monitored down.
:-) Where better than in a hostelry which invariably has always wenches and benches.
"Assume amanfromMars posted from the Rose and Crown" .... Jim 59 Posted Monday 15th August 2011 16:16 GMT
Yes, please, it is a remarkably effective cover with many attractions. And so hard to believe that anything remarkable can be developed therein ...... but to those in the know, is it most probably considered hard to better for a smart beta, Jim 59.
spinning vs flash
I don't really care as long as it's no slower than I already have, is 4-5 times the capacity I have (1PB or so), uses less power and is cheaper.
Ease of migration helps a lot, but LVM pv migration seems to work fairly well...
Having said that:
Flash drives are set to hit 1.6Tb by year end. How big will they be be by end of 2012? How large will spinning rust be? How fast will they develop bad blocks? (both of 'em). How much vibration will they handle? (this is critical in disk arrays, 48 drive drawers need quite a bit of design to keep from reducing disk life).
Now does anyone want to buy an old DEC 2.2Tb array? (3 * 18u high cabinets, around 4kW power draw...)
Most datacentre admins choose from what's available at the time they need to buy. Not what might be available in a year's time.