The Hot Aisle Logo
Fresh Thinking on IT Operations for 100,000 Industry Executives
« Previous: Technology Predictions for 2012
Next: Solid State Storage changes everything »
Published by on Monday 6th February 2012, 15:54 | Related | Filed Under

I just finished reading EMC company blogger, Chuck Hollis article about EMC VFCache, a server side flash storage technology that competes with the Fusion-io, ioDrive head on.  Most interestingly, the product name is chosen carefully to ensure customers don’t get confused that VFCache might be designed to replace the normal disk based EMC storage array controller products.  In fact VFCache acts as a caching layer between a back end storage controller and a front end server demanding massive IOPs. The VFCache is integrated via a filtering driver that reduces the load on the back end disk based array whilst maintaining and managing cache concurrency between the VFCache and back end.

EMC call VFCache, “Server-side flash storage cards that integrate with the rest of the extended storage environment” – neatly covering off one of the major objections to PCI card flash storage for mission critical applications, that is, what do we do about resilience, hardware failure and disaster recovery? Acording to Chris Mellor at The Register“The EMC cache increases 4KB – 64KB block random read I/O speed but not write I/O speed.  VFCache will not cache read I/Os larger than 64KB. There is no write caching.”  So we can expect VFCache to deal with intense write activity quite badly, where the back end storage is not able to keep up with the front end cache.  A risk of imbalanced systems I sense.

Dedicated, solid state only storage offers quite significant advantages for the consumer of high IOPs applications, like SAP, Exchange Server, SQL Server, Oracle etc..  Quite dramatic increases in read and write performance combined with extreme low latency drive application performance to new levels whilst simultaneously reducing the quantity of hardware (and therefore application licences) needed to run them. Good to see that EMC are waking up and recognising the value of engineered solid state storage products even if they are (unsurprisingly) still strongly wedded to their core disk business.  A much better approach is to recognise that the storage world is moving inextricably towards flash and trash, Solid State for performance and SATA disk for capacity.  Prices are falling rapidly as flash fabrication plants come online and it won’t be long before high spin speed disks are totally dead replaced by their much better performing SSDs.

Share
  • Chuck Hollis

    Hi — thanks for the post, and the shout-out.

    Two point of clarification, if it might help.

    First, it’s a bad idea to cache persistent writes on *any* server-side storage device — cache, disk, etc.  Lose the server, lose the data — in addition to defeating just about any server clustering technology out there, including the popular VMotion.

    So persistent writes will have to be written through to some sort of external storage device that protects the data against node failure, in addition to exposing written data to other nodes.

    Not all written data is persistent — think tmp files, etc.  So VFCache can also be configured in a write caching mode if that’s the case — none of the writes on a logical device so configured will be written through to the array.

    Finally, I do want to point out that EMC’s block arrays land each and every write in non-volatile cache — array-side, yes — but a not insignificant performance boost.

    Thanks again!

    – Chuck