I ran into Woody Hutsell, EVP at Texas Memory Systems, last week. He graciously agreed to a talk on camera about their experience with flash and DRAM-based solid state storage.
TMS sells both: a DRAM-based SSD with multiple FC and Infiniband ports; and a 2 TB flash box with 128 GB of DRAM cache. Woody offered some interesting insights. For example, workloads with a large number of writes – even if they are a small percentage of the total workload – may not be suitable for flash-based storage.
Here’s the video:
Blame me for the shaky camera work.
Disclosure: I taped and edited this gratis.
Comments welcome, as always. BTW, Google now accepts files up to 1 GB. Seagate and WD should be happy.
Woody Hutsell said: “[Our] RAM-based SSD is going to run about $650/GB.”
Hmm let’s see I have the choice of buying his device at $650/GB, with a max throughput of 4 GB/s per link (4X DDR InfiniBand, FC would be less performant), and a latency of at least 1000+ ns (just counting the InfiniBand latency, not the additional overhead imposed by their device).
Or I could run my app on a UPS-backed regular server with 128 GB of DDR2 667 ECC registered RAM [1] at 1/18th the price/GB [2], 2.7x the throughput [3], 1/5th the latency [4]. And create a RAM-drive or let the OS intelligently use the RAM as the filesystem buffercache (Linux is very good at it).
Please someone give me a good reason I shouldn’t go into this business and make big bucks by selling RAM at 18x its street price.
[1] Like this 8 sockets, 64 DIMM slots, Opteron server: http://www.sun.com/servers/x64/x4600/specs.xml
[2] $36/GB for DDR2 667 ECC registered 2 GB sticks: http://www.newegg.com/Product/Product.aspx?Item=N82E16820134328
[3] 10.7 GB/s for dual-channel DDR667 (667 (MT/s) * 128 (bits) / 8 (bits/bytes))
[4] ~200 ns of average latency in 4 and 8-socket HT configurations
That’s why so many SDD firms have gone under over the years. It is a hard sell. I know. I took a crack at it back in the early 90s.
And also why an SSD at $112/GB is a big deal.
Network-based memory has advantages over a single system – you spread the cost and the benefits over more systems.
Also, IIRC, Infiniband latency is
quite a bitless than 1msµs, but I haven’t looked at it for a few years.Robin
Agreed.
About InfiniBand latency, you didn’t read me correctly: I was talking about 1000 ns (or 1us), not 1 ms.
PTZ – you’re correct. Oops. Corrected the comment.
Robin