Up until now flash memory was for thumb drives and trying to speed up slow notebooks suffering Vista bloat, which has been unsuccessful. Adding flash cache to hard drives also proved disappointing in getting a performance edge. Top-end transaction-processing businesses don't use flash. They use military-grade DRAM solid state disks (SSD) instead, which are very expensive.
Back in September I wrote about the imminent arrival of flash memory-based SSDs with TMS' DRAM cache front-ended RamSan-500 leading the charge. That was still quite expensive. Now EMC and HP have upset the apple cart with flash SSD products and opened the door to widespread adoption of flash. There's an opportunity here for fast-moving storage drive array resellers.
EMC has announced a flash SSD version of its top-end Symmetrix DXMX-4: the Enterprise Flash drives single-level cell STEC NAND flash arrays with 73 or 146GB capacities. The idea is to use them for database tables or transaction data with the fastest possible response time where the user can not afford more expensive DRAM-based SSDs. The cost is about 30 times more per megabyte than fast Fibre Channel disks. Each flash drive delivers, says EMC, about the same I/O bandwidth as 30disk drives.
So now every other drive array vendor is scrabbling to try and catch up. Obviously STEC is a potential supplier but so too are Micron and SanDisk.
HP is using SanDisk flash drives in a new and very energy-efficient PC (currently for the US only) called the dc7800. It uses a 16GB SanDisk flash SSD making it about $300 costlier than a normal disk drive dc7800. SanDisk also makes 32GB and larger SSDs. Apple's latest MacBook Air also has a flash drive option. Suddenly flash memory is cool and almost affordable.
The drive array manufacturers will take time to respond, especially ones with lots of storage middleware that will need functionality validated against flash drives. Hitachi Data Systems is already signalling a slow response, to wait and see if there is a real need.
As long as a flash drive has a SATA/SAS interface, has wear-levelling algorithms to get over the write cycle limitations, and is certified for use with SATA/SAS controllers, there is no real problem. It means that a 'white box' storage array manufacturer can add a set of flash drives to their product line and offer three tiers of storage: tier 0 performance-centric flash; tier 1 middling performance SAS; and tier 2 capacity-centric SAS. The tier 0 product will outperform any competing manufacturer's Fibre Channel or SCSI drives, no matter how fast they are spinning. It uses a whole lot less electricity too.
Imagine the sales pitch: "Look at my flash array product. It outperforms any HDS, HP, IBM, NetApp or Sun disk drive array, costs half the price, and uses less energy." That would make HDS, HP and IBM go green - with envy.
You can go to customers and say the largest single contributor generally to slower application performance is disk drive latency. Let's get rid of it, just like that.
There seems to be margin headroom with EMC's DMX-4 flash drive pricing so there is time here, six to 18 months perhaps, before all the major drive array vendors have their flash-based products announced.
Where would you get flash drives from? EMC and HP have validated STEC and SanDisk as flash drive suppliers. Micron will be keen to demonstrate its flash drive potential and Intel will be looking at the space as well.
There is another angle here. If a customer is using de-duplication software, a flash-boosted storage array will run the software faster. I do not think back-up to disk will be boosted unless the data fits in the flash storage. If it does, you could back up to flash and migrate the data to SATA drives, leaving the flash empty for the next back-up.
This is a short-term capacity problem. In a year flash capacities will have doubled and the cost halved. But by then the main drive array vendors will have their flash-enabled products too.
Next year we should see 128GB flash drives and then 256GB ones and we'll be looking at the potential wholescale takeover of fast Fibre Channel disk by flash drives. These drives could also start to limit fast SAS drive sales.
One more thing. Seagate and others are encouraging the use of 2.5-inch drives instead of 3.5-inch ones in storage arrays as a way of getting more I/Os out of a drive shelf. There are more spindles in a box populated with 2.5-inch disks than 3.5-inch ones.
The flash drive invasion will put a stop to this. No matter how many small disk drives you put in a drive array shelf the I/O total won't even approach flash-drive shelf levels. Also, as flash-drive capacity becomes cheaper and cheaper it is possible the 2.5-inch drive array market will wither and die.
Flash drives are like a combination Toyota Prius and Porsche 911, being both greener and faster than Ford Mondeos or BMWs. If you can react fast enough you could turn your Seat or Skoda, even your Trabant into a Porsche-plus-Prius combo and deal a satisfying flash, bang, wallop to your competitors.