Full-stack Philosophies

James Morle's Blog

RSS Feed

Death of the Storage Array

Posted on 3:03 pm August 19, 2013 by James Morle

When I wrote this article for The Register in October 2010, there was a torrent of naysayers and witch hunters spewing their opinions in the comments section. I don't have a problem with that, I was only expressing an opinion myself, after all. I don't actually own a time machine and so any of my predictions are only based upon observations in the industry. It is heartening, though, to see that the prediction is coming true a little bit sooner than I expected with the announcement from EMC that VMAX will be relegated to the Capacity Tier. At least it seems that EMC 'get it', and are doing the right thing.

Like it or not, solid-state storage is the future of high-performance storage, and the hardware/software architectures have to change to keep up with the dramatic reduction in latency solid-state storage brings. Flash storage isn't perfect, but the next technologies in solid-state storage remove many of its deficiencies and retain or improve on those latency numbers. There is no going back, only forwards.


6 comments on “Death of the Storage Array

  1. >solid-state storage is the future of high-performance storage, and the hardware/software architectures have to change to keep up with the dramatic reduction in latency

    ...Good points, but I would change the above to read "Non-Volatile Memory is the future..."

    ...Lines will soon become exceedingly blurry between what we think of as "memory" and "storage".

      • "architectures" won't have to change. The changes will be intrinsic. Apps won't even know (or need to know perhaps) whether a page is in NVM or DRAM. Intel has patented 2LM. I can't say anything more than what's publicly available on the matter, of course, but if you read the patent and take in to consideration the fact that the patent is only showing certain embodiments of 2LM you can start to read the tea leaves http://www.sumobrain.com/patents/wipo/Two-level-system-main-memory/WO2012087471.html

        The short of it all is that DRAM's future is cache, not memory.

        • Well you are certainly privy to more futures than I, I will grant you that! But is the inclusion of an additional tier of cache, whether controlled by hardware or software, not an architectural change? The move from read(2)-and-write(2)-alike system calls to user mode loads and stores? Seems architectural to me...

          • libC/libaio applications will still be able to benefit from 2LM *without a single line of code change*. That's all I'm talking about in this thread.

            Now, a scalpel job to, for instance, retrofit from read/write to load/store via a Flash vendor API (ala FIO, EMC) is a step along the way to the future of non-volatile memory... but that is another discussion...the two threads are related by separate.

  2. Folks continue to think in terms of storage being a "device". It isn't anymore. Like everything else, it must be virtualized. Who cares if it's on flash, ssd, spinning rust, spooled or whatever? All a data centre manager (cloud or private is immaterial) wants to know is there is some storage, somewhere, with a given performance level. Vsphere and VSA is a step in the right direction and it'll only expand from there.

Leave a Reply