FusionIO

Anyone heard of Fusion-IO?  I’ve been doing some reading and it seems like if the press is reliable they may have hit that breakthrough I’ve been looking for.

Essentially it’s up to (currently) 320GB of NAND-Flash storage memory that fits into a PCI-E x4 slot.  Booting from this is supposed to come sometime in Q3, but when it does, watch out.  The numbers they are quoting:

  • Over 125,000 IOPS of random sustained read/write packets
  • 700MB per second of bandwidth – Sustained

make them easily one of the best performers in the game.  You can transfer 4.5GB (a single-sided DVD) in about 5 seconds.

The fact that they have a 320GB card out, and a 640G card due later this year, means they’ve got conventional SSD beat.  The only serious drawback I can see is the lack of a 2.5″ form-factor SATA interface for laptops.  But who knows – that may also be coming down the pipeline soon.  I’d replace the 80GB SATA-II drive in my Latitude D620 with one of the 80G ioDrives in a heartbeat.

Check them out at www.fusionio.com and let me know what you think.

4 comments

Skip to comment form

    • on April 24, 2008 at 12:32 am
    • Reply

    Saw them at SNW in Orlando a little while ago. Had three ioDrives running on a gaming machine; not sure what capacity; I believe it was a software RAID. Showed 2+GBs per second streaming, and well over 200,000 IOPS. Amazing. Also, each ioDrive only consumes 6 watts, so an extremely GREEN device. I want one!!!

  1. yeah – the funny part is – I run Vista on my PC at home, with moderate success, of course I have as my disk back-end right now a 4-drive Raid-1/0 set so there is a certain amount of performance benefit there. I can still watch the harddrive thrash when i do something overly complicated—-like logging in.

    This would end that struggle once and for all. 😉 probably a 15 second boot time. 🙂

    • on April 29, 2008 at 6:52 pm
    • Reply

    Yes, yes I have. This was the first question I brought up when everyone was ranting and raving about the flash disks in EMC’s arrays for those one or two database servers that need maximal performance. This will kick the crap out of that any day of the week.

  2. Yeah – My only real question of course is whether they have fond a way around the 100,000 write-erase-cycle limit for NAND flash memory.

    Most flash-based systems use something called write-levelling, which is largely a way of randomizing IO (in flash memory, random IO is not a performance problem like it is in hard-drives) so that each block gets eat an equal amount of time.

    NAND flash memory has a limited number of write-erase cycles, around 100,000 under current design. Now a write-erase only occurs when the blocks need to be set to zero. IE you can change 0000 to 0001 to 0011 to 0111 to 1111 without going through an erase cycle, (or any combination thereof) however in order to change any bit from 1 to 0 requires that that entire block be erased and re-written. This could prove to be a substantial limitation for the type of storage.

    That being said, it’s just as much a limitation for every type of NAND-Flash based storage. If Fusion-IO has come up with a solid write-levelling algorithm and can truly distribute the Io across the entire card, then they by far have the superior product.

    As I understand it, most commercially available SSD disks write as a normal disk, which could cause them to run up against that write-erase barrier much sooner. (IE if a disk is 30% full, the first 30% of the drive is most active and will hit the limit far before the end of the address-space.

    Assuming of course that the SSD manufacturers haven’t already adjusted to this with a write-levelling algorithm of their own.

Leave a Reply

Your email address will not be published.