Viking: Why Wait for Nonvolatile DRAM?

Viking NV-DIMM and Supermicro serverViking Technology on Wednesday announced that the company has arranged with SuperMicro to deliver solutions that should solve issues key to flash storage array designers.  Supermicro will ship server boards that have all the hooks necessary to support Viking’s nonvolatile DDR3 DIMMs without users having to wait for this kind of support to arrive in servers based on the standard Intel platform.

Viking’s marketing folks tell me that NV-DIMM support is something that its customers have been asking for, and that it will be a standard component of Intel’s Haswell platform.  Trouble is, Intel won’t be rolling out server designs based on the Haswell platform until late next year, and Viking’s customers want NV-DIMM support now.

Rather than wait for Intel support, Viking has taken the initiative and ported a Viking-developed BIOS and hardware solutions to Supermicro, the vendor of choice for a number of Viking’s larger customers.  This will allow these customers to purchase an off-the-shelf solution to a problem that they have so far had to solve on their own using internal designs.

Why don’t these customers just use PCIe flash for this problem?  It has a lot to do with bandwidth.  The application NV-DIMMs are most often used for is caching and journaling for SSD arrays, and this a a very high-bandwidth application.  A DDR3 DIMM supports a bandwidth of 12GB/s, while one PCIe Gen2 lane is 0.5GB/s.  The highest performing PCIe SSDs use 8 such lanes, peaking out at 4GB/s, or 33% of the bandwidth available from a single NV-DIMM.  Another problem is that these applications use the NV-DIMM as a write log, which is a high-write application.  A write log presents wear problems to flash, so DRAM is preferable.

Viking’s NV-DIMMs are based on standard DDR3 DRAM, and have internal back-up flash that copies the contents of the DRAM in the event of a power failure.  The proprietary BIOS hooks move the processor’s internal state to the NV-DIMM, allowing the server to re-boot in milliseconds, rather than seconds, when power recovers from a failure.

It’s impressive stuff.  Hats off to Viking for making it happen today, rather than waiting for Intel’s Haswell support.



4 thoughts on “Viking: Why Wait for Nonvolatile DRAM?”

  1. Jim, I’m curious as to why SSDKeeper hasn’t been mentioned. If DRAM access is 1,000 times faster than SSD access, and I have extra DRAM, why not put it to use?

    Honestly, I did try the SSDKeeper Trial-version and saw a 49% reduction in SSD accesses (about an hour of I/O, over a 2-week period). Am I missing something, like the risk of loosing data during a power-failure?

    These guys (Condusive Technologies) are surely flushing that DRAM cache to SSD, periodically. Any idea about the safety of Diskeeper 16 and SSDkeeper 1.0? Both support DRAM caching of SSD’s and 16 also defragments HDD’s.

    Thanks for your insight and heads-up information.

    1. Michael,

      Although I haven’t tried SSDkeeper myself, your testimonial certainly makes it sound like it’s worth trying. I will contact them.

      There’s a lot of improvement that still needs to be done to make software make the best use of SSDs. i guess that SSDkeeper has helped in that direction.

      Thanks for the comment.


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.