Hard drives simply can't meet the intense performance demands of cloud apps, claims Dave...

Hard drives simply can't meet the intense performance demands of cloud apps, claims Dave Rosenberg, the CEO of a cloud management software firm. Only flash memory is up to the task, in his view, even if costs remain a concern. Fusion-io's (FIO) flash modules help run cloud apps for Apple and Salesforce.com; flash vendors OCZ, STEC, and SNDK are also targeting this space. (previously)

Comments (2)
  • Conventional Wisdumb
    , contributor
    Comments (1800) | Send Message
    That's actually quite a profound insight with awesome implications if true.


    NetApp one of my faves would be buggy-whipped if this were true long term.


    "Industry giants EMC and NetApp have begun to offer "flash as cache" solutions that can be bolted on to their existing products, though critics point to two significant drawbacks with this approach: adding flash to existing solutions can be incredibly expensive; and their function is limited to a read cache."


    The second limitation is critically important, according to Ed Lee, an original member of the Berkeley RAID team in the 1990s who is now lead architect at Tintri. "With a bolt-on flash as cache, you're lucky if 30 percent of the application traffic benefits from flash. Basically, you're still getting disk-level rather than flash-level performance," Lee said in an e-mail interview."


    Read more: http://cnet.co/A23hXl
    Read more: http://cnet.co/xuZ0me
    15 Feb 2012, 09:14 PM Reply Like
  • davcnslt
    , contributor
    Comments (140) | Send Message
    To add a bit more insight to your excellent note:
    In an article today, a Univ Of Calf @ San Diego research project was discussed in which it was stated:


    "While the density of SSDs grows and the cost per gigabyte
    shrinks, "everything else about them is poised to get worse,"


    "This makes the future of SSDs cloudy: While the growing
    capacity of SSDs and high IOPS rates will make them attractive
    for many applications, the reduction in performance that is
    necessary to increase capacity while keeping costs in check
    may make it difficult for SSDs to scale as a viable technology
    for some applications,"
    - Laura Grupp, UCSD researcher & spokesperson


    To summarize the article: As flash denizens pursue smaller and smaller line widths in the die, in order to deliver higher discrete data per device (density) and become more competitive with magnetic recording (disk drives) pricing (where current pricing in like-for-like capacity devices is >10:1, the performance falls off to a point where "write speed for pages in a flash block suffered dramatic and predictable variations in latency".


    Flash has been hailed by many observers as the solution to disk drive (HDD) latency causing diminished input/output operations per second (iops). iops in HDDs, as has rightfully been pointed out by these observers, suffers due to the seek time from mechanical track-to-track seeks required to position the read-write head accurately over the data. The difference, according to this research, is that flash latency will grow over time and is largely due to errors in which data will have to be recovered and re-mapped in the flash device.


    The speed of iops achieved with flash memory is well established and improved performance in enterprise storage systems, particularly those operating in a virtualized system, has long been a primary rationale for selection of flash memory over hard disk drives. I would not choose to argue with Mr. Rosenberg on this point, but depending on where the diminishing returns begin to occur (is it accute in today's avg line widths of 25nm or is it acute at some future implemenation?) there is a major point suggested that could dramatically alter the fortunes of data storage strategist in the relatively near future.


    I'd hasten to add that there are many questions left unanswered by this research and readers should remember that the assumptions made are based on mathematical extrapolations that may or not be challengable by consideration for other influences such as the technology roadmap for photo-lithography, error correction codes, and evolving data compression techniques. We can expect to see much on this topic in ensuing articles, blogs, etc.


    In the meantime readers may want to consider recent research performed by an international team led by the University of York's physics department which can be read at http://bit.ly/z5EZCi, as well as recent research announced by IBM at http://bit.ly/yBpV3d


    Disk drives are very far from their end of life in my opinion, though drumbeaters for solid state drives such as Mr. Rosenberg would have you think so.
    17 Feb 2012, 03:35 AM Reply Like
DJIA (DIA) S&P 500 (SPY)
ETF Screener: Search and filter by asset class, strategy, theme, performance, yield, and much more
ETF Performance: View ETF performance across key asset classes and investing themes
ETF Investing Guide: Learn how to build and manage a well-diversified, low cost ETF portfolio
ETF Selector: An explanation of how to select and use ETFs