The V-Max dust is settling
How will EMC's competition respond?
Comment EMC raised a marketing whirlwind with its highly scalable Symmetrix V-Max last month, and much dust was raised as competitors and EMC made claims and counter-claims. What will EMC competitors actually do, though, as they assimilate the impact of V-Max?
It's possible to discern likely steps forward in enterprise array design by stepping back a moment, roughly a decade or so, over-simplifying and saying that the enterprise storage array space was then dominated by two main storage architectures. There were modular arrays, like EMC's Clariion, NetApp's FAS products and HP's EVA, with other products from IBM and HDS, and there were monolithic arrays with more powerful controllers and greater performance and scalability.
Then second-generation SAN array architectures came along, mostly based on commodity processor technology. Compellent offered block-level tracking and automated the placement of blocks based on activity levels. This was a new kind of modular array. Pillar built its Axiom with a 2-level controller design and a third storage node building block. This enabled Pillar to scale performance and capacity separately and neatly and develop its application-aware, differentiated quality of storage service.
3PAR developed its enterprise utility storage array using multiple controllers connected in a type of cluster, using a proprietary ASIC-driven interconnect that linked them and the storage shelves. There can be eight such controllers in its InServe product.
The idea of separating controllers from the storage is one of the main aspects of these designs, and one that Hitachi took on board when it introduced its USP high-end product in 2004. The virtualising USP-V controller was introduced in 2006, with IBM introducing its in-fabric SAN Volume Controller in that decade too. Both of these meant that you could virtualise a huge amount of storage behind the controller, and develop and scale the controller technology separately from the storage node or enclosures, which could have their own performance and capacity scaled as well.
We can view Symm V-Max as EMC realising that modularisation, combined with scalable controller technology, can deliver a better and more cost-effective way of developing array controller functionality and performance, and scaling controller and storage node performance and capacity independently. V-Max can have up to 8 clustered controller nodes currently, the 3PAR number, but is set to go higher. By clustering controllers you can scale array controller performance higher and faster than by riding, say, Intel's processor development curve.
The main Symm V-Max competitors are 3PAR with its InServe, HDS with its USP-V (OEM'd by HP and Sun), IBM with its DS8000, and NetApp with its FAS 6000 product line. How are these likely to develop?
It seems probable that all four will develop clustered array controllers using Intel Nehalem 5500 building blocks. These quad-core CPUs are powerful, available in multi-socket configurations, and with 6- and 8-core versions coming that there is simply no need to use anything else. Opteron's maybe, or Power6 in IBM's case, but for everyone else, Intel rules.
Such controller power is needed to provide the virtualised, multiple storage tiers, from SSD to SATA capacity drives. These will feature thin provisioning, some form of dynamic data movement across tiers, data protection based on snapshots, unified block and file access, increased access ports (consequent on server virtualisation and cloud-style computing), increased back-end ports (ditto), and possible array partitioning into different storage personalities such as online, nearline, and archive.
There will need to be a cluster interconnect, either proprietary, InfiniBand or EMC's RapidIO, but not Ethernet as RDMA over Ethernet isn't ready yet.
Next page: The competition's responses
Fujitsu seems to be overlooked a lot, I was going over their site a few days ago and was quite surprised to see this monster array which at least on paper seems on par with many features that other high end vendors offer, and one or two unique ones.
Up to 8 controllers, and 2700 disks
The Cyclic cache mirroring is pretty unique I think. I sent a note to my 3PAR SE who was pretty impressed as well though says they don't encounter Fujitsu much here in the US, must be more popular in Europe. With such an impressive system I'm surprised they don't push it more in the US.
It doesn't appear that they do block based virtualization like 3PAR or Compellent. Also I don't see an indication of supporting hyper-dense drive enclosures. 3PAR's S/T class support 40 drives in 4U. Providing such density is fairly complicated as you need to have software that can automatically lay data out in a way to handle multiple drive failures(to replace a drive in a 3PAR S/T class you must remove 4 drives). Another big vendor that was pitching their stuff to me said they were going to go hyper-dense this year as well though I haven't seen it happen yet.
Also I think the USP (non USP-V) is also a virtualizing storage controller. the USP-V is just the low end model of it.
I think the V-MAX would of been more impressive had they scaled the architecture down to some extent so it's more affordable. Maybe they will do that in the future. I saw a price somewhere listing a V-MAX entry level system at $250,000 with no disks, the article didn't mention what it included exactly for that price. I suspect their interconnect is what drives a lot of the price of the controllers.
When 3PAR announced their first array back in 2002 it was advertised as something similar - 8 controllers, 2560 drives, 192 fiber ports, 28 gigabytes/second interconnect on the 8-node box. Entry level pricing starting at $100k - http://www.3par.com/news_events/press_releases/20020610.html
I just thought of motorbikes.
Maybe I've missed the point, but aren't Symmetrix arrays already modular? You add director cards as you need them, drives as you need them and enclosures as you need them. I assume I've missed something here...