Fusion-io bubbles away but will it erupt?
Getting into a lava about server attached SSD
Comment Market seismometers are detecting moves from Fusion-io, the server-attach solid state disk startup. Is it about to erupt and deliver on its potential?
Fresh from a new funding round and led by a new CEO, David Bradford, Fusion has hired a new SVP for worldwide sales, Jim Dawson, who comes to the post after five years at 3PAR: five very good years, judging by 3PAR's trading history and growth. Only a slight stumble in its last reported quarter took it into loss. Has Dawson paid the price for that slight hiccup, which marred 3PAR's earnings report? He's been welcomed with open arms at Fusion-io, especially as his move from a successful start-up selling to enterprise customers implies he sees a great future for Fusion.
Dawson appears to have replaced Dixon Doll, a previous Fusion sales VP, although Fusion is not saying. His job is to drive what's called an aggressive worldwide sales expansion. His previous employer, 3PAR, has a solid state drive (SSD) strategy, one of adding SSDs as a replacement for hard drives. This is the commonest tactic employed by storage array suppliers and follows on from EMC's lead.
He pours cold water on this idea, saying: "With the seemingly insatiable demand for storage performance driven by many applications, databases and data center virtualisation projects, it is nearly impossible for traditional, disk-based storage infrastructure to continue to scale in performance. Simply putting solid-state disks within the array’s architecture doesn’t solve this problem. But Fusion-io’s innovative approach to system acceleration does by placing the solid-state technology in the server where the applications reside while increasing performance by several orders of magnitude."
Well, he would say that, wouldn't he? Fusion isn't alone with its idea of caching servers. Sun has used Intel read and write flash caches to accelerate its servers. NetApp is using a server caching - actually controller caching - approach to increasing I/O performance, with its DRAM-based PAM (Performance Acceleration Module.) It is, perhaps, a bit of a conundrum why NetApp and Fusion-io aren't working together in this area, NetApp having a NAND-based PAM in the works.
Fusion-io has a recent history of eye-catching initiatives, ranging from Apple co-founder Steve Wozniak's appointment as Chief Scientist, to a demo with HP of a million plus IOPS from flash memory-based ioDrive PCIe cards plugged into an HP server. This followed on from an earlier demo of a million IOPS from an IBM SAN Volume Controller using the same ioDrive technology. Michael Dell is also an investor in Fusion, though Dell has yet to announce any use of Fusion's technology. There's lots of potential credibility here.
HP is now shipping ioDrive-based IO Accelerator cards for its blade servers. There is a new UK-based channel partner, Diamond Point International, which will sell the ioDrive and ioDrive Duo (effectively two ioDrives on one card) to system integrators and independent hardware vendors needing to add a fast, non-volatile memory resource to servers.
The ioDrive uses single level cell (SLC) NAND in its 80 and 160GB manifestations and multi-level cell (MLC) NAND in the 320GB version, thought to be 2 bits per cell. The ioDrive Duo is an SLC product for its 160 and 320GB models and MLC for the 640GB model. The Diamond Point announcement said a doubled capacity ioDrive Duo, one with 1.28TB capacity, would appear in the second half of this year. This implies that a doubled capacity ioDrive, one with 640GB would appear at the same time. Both will probably be MLC technology products
Fusion is also working on ioSAN technology. The ioDrive is direct-attached storage for a server. It pairs single level cell NAND flash chips - MLC for the highest capacity model - with a Fusion controller and PCIe interface on a single card. If that server is then given a 10GbitE or 40Gbit/s InfiniBand interface it can make this flash memory resource sharable with other servers. Fusion says this ioSAN can be based on any vanilla server and, with hard drives added, that server can despatch fast access data from the flash memory and ordinary access data from its bulk storage hard drives.
It is, in effect, a low-cost SAN with both high-performance and bulk capacity. Fusion is talking of building systems that can do millions of IOPS and have tens of gigabytes of sustained bandwidth, with less than one millisecond of latency. On the face of it this stuff looks great for a unified data centre with a virtualised and bladed server environment. Depending on the price, it could be a viable alternative to an FCoE-capable SSD plus HDD storage array. Just use a spare virtual machine in a blade server to host the ioSAN and then ask: "Who needs a storage array?" Is it really that simple, though?
This is not a tactic designed to attract storage array vendor support, which may explain the lack of any storage array supplier partnerships.
Fusion is working with Samsung on the ioSAN development, and obviously going to use Samsung NAND chips. It is pitching this as a stage in a transition from server-attached to server-deployed, network-attached storage. It says its ioSAN represents the first step in this evolution and has been demonstrating it with Samsung’s NAND flash technology at Interop 2009 in Las Vegas. There is a possibility that the flash capacities needed for a shared resource may require a move to 3X MLC (3 bits per cell) and clever technology may well be needed to provide the read/write access speeds and working life needed. Is that why the partnership with Samsung has proved necessary?
Fusion-ion has demo'd big, recruited glamorous IT celebrity support, and kept up an energetic sense of momentum up to and after its CEO change, but will it deliver? Will the server vendors adopt its technology, with Dell, Fujitsu, HP, IBM and Sun servers using its products as standard? If that happens, if a global server vendor adopts Fusion-io technology, then this potential volcano will erupt.
On the other hand, if Fusion's future is with many smaller sales through IHV-based channels then the sudden eruption may not happen. Instead, we'll see a more restrained flow. Which type of Fusion is going to evolve should become clearer in the next few months. ®
Sponsored: Benefits from the lessons learned in HPC