When you read the Fusion-I/O Specs, they have an access time of 50 microseconds, and 95,000 IOPS for the 160GB SLC ioDrive; that’s over 100 times faster than a traditional disk drive. Look at EMC’s assessment of the potential of flash and you see a more sober assessment of 10-30 times faster. So is Fusion-I/O a much better product and set to dominate the market?
When you look in detail at the spec of the Zeus IOPS 3.5 inch SSD from STEC which is used in the EMC storage arrays the difference between the two technology’s are minimal. Clearly the difference in performance comes from the overhead of running the SSD in a storage array. Is using a disk-drive form factor for storage a wrong decision?
I would argue that Fusion-IO have a more difficult road to market. Disk drives have one essential characteristic that differentiate them from the rest of the system – once written and acknowledged, the data is considered safe from any power or other failure. The ecosystem of system’s software, storage controller software and applications have been written that protect disks, manage utilization, virtualize them, share them on SANs between servers, copy them, back them up locally and remotely, restore them and optimize performance. Sure, the current storage controller arrays need to be adapted to cope with much lower latencies and higher performance of Flash SSDs, but that will take one generation of storage controller to achieve by most vendors.
The Fusion-IO products interface is PCI-Express x4. Very flexible, but you have to start from scratch for software support. It is easy to write a driver for it on a PC, but much harder to re-write the storage infrastructure of software that has been developed to support the only common medium that is safe after it is written. Even on a PC, the flash memory stick is treated like a disk drive.
In a recent article on Wikibon, I looked at the projects for flash storage compared with high-performance FC drives. The projections indicate that within three years, high-performance drives will not be competitive against flash drives. The storage architecture will revolve round and mixture of high-performance flash drives for IO activity, and serial ATA drives for bulk data storage. In three years we can expect new storage controllers. Three years is far too short a time to develop an effective replacement storage ecosystem.
My bet is that Fusion-IO will have great success in niche enterprise markets, and will have a role to play in some server architectures. My bet is that they will need to embrace the disk drive form factor to be a player in enterprise storage. EMC and STEC have the right vision for enterprise storage.