snw_largeOn day 2 of SNWE, I popped down to the Fusion-IO stand.  At the IP Expo event earlier this month, I was initially told I couldn’t video my chat with on of the presenters on the stand.  Fortunately that appears to have been no more than a mixup, and Mat Young (marketing guy in EMEA) happily did a walkthrough of their extravagant showcase – a Fusion-IO card running a massive video wall of 1800 consecutive videos and 400+ desktops.  The video is attached here.

Fusion-IO (if you’re not familiar with the company) produce a number of plug-in cards that emulate hard disk drives.  They offer ultra-low disk I/O latency and high performance as the card is directly on the server bus (via a PCI-Express connection).  Where other companies have focused on speeding up the array (via SSD) or even building dedicated SSD arrays, Fusion-IO have taken a different approach and moved the disk back into the server.  There are plenty of use cases where this is a good idea, including server and desktop virtualisation.

Now take the existing integrated storage infrastructures, with server, network and storage.  Imagine replacing some of the storage with a few Fusion-IO cards, that let you drive the servers at a much higher utilisation.  Data that isn’t actively used gets moved off to SATA drives; this doesn’t have to be implemented with an expensive array as the performance isn’t needed.  Instead, a much lower tier product could be used.  Of course this would require an intelligent hypervisor capable of managing the data placement, but that’s not difficult to achieve.   The hypervisor already knows about which disk blocks are most active.  It requires only the design of a tiered VMFS.  If Fusion-IO are successful, the unified computing stacks of the future could be very different from those we see today.

Tagged with:
 
  • Roger Weeks

    I think FusionIO is a great technology, but I think any application which keeps the data local to a single server has an issue with scaling. I agree that scaling certain kinds of applications across hundreds of servers each with their own local storage is possible – look at Google for example.

    But I don’t believe that this type of application-based architecture is applicable to all kinds of computing. You certainly wouldn’t want to use this for unstructured file storage, for example. Nor would I use it for a transaction-oriended SQL database.

  • http://twitter.com/simspot Andrew Sim

    @Roger,

    I hear what you are saying about scaling out across hundreds of servers, but I wouldn’t rule out using it for traditional OTLP databases. Recently, we’ve been in talks to re-architect an Oracle IO-bound application with a solution that employs Oracle 11gR2 Flash Cache sitting on Fusion-IO cards in a multi-node RAC cluster. We’re pretty confident that any IO-bound application will perform significantly faster with this setup. I expect other database vendors to introduce features similar to ‘Flash Cache’ as this faster tier of storage will become more predominate in the next few years.

Looking for something?

Use the form below to search the site:


Still not finding what you're looking for? Drop a comment on a post or contact us so we can take care of it!

Visit our friends!

A few highly recommended friends...