turning siFlash past 10 (GB/s that is) …

Yeah, that title is a Spinal Tap homage.

We are bringing a new siFlash unit to the HPC on Wall Street conference. This uses our new chassis, an updated kernel, and lots of tuning. Still have much more work to do … but its probably good enough to ship now.

I ran a few quick speed drills on it. 4.2 GB/s streaming write , and 10.7 GB/s streaming read with 96 simultaneous processes.

No, I am not kidding.

Run status group 0 (all jobs):
   READ: io=381312MB, aggrb=<strong>10680MB/s</strong>, minb=<strong>10680MB/s</strong>, maxb=<strong>10680MB/s</strong>, mint=35703msec, maxt=<strong>35703msec</strong>

Thats reading 1/3 of a Terabyte in about 36 seconds. So 1TB ought to take about 100 seconds or so.

Writing 1TB takes about 287s

Run status group 0 (all jobs):
  WRITE: io=1140.4GB, aggrb=4067.9MB/s, minb=4067.9MB/s, maxb=4067.9MB/s, mint=287067msec, maxt=287067msec

Reading is a little faster.

Run status group 0 (all jobs):
   READ: io=1140.4GB, aggrb=<strong>11074MB/s</strong>, minb=11074MB/s, maxb=11074MB/s, mint=105449msec, maxt=105449msec

Thats 1TB (really 1.14TB) read in 106 seconds. 11.1 GB/s.

This is one of the fastest self contained storage boxes on the market.

We’ve got to rebuild some of the raids and do a better job on interrupt management. This isn’t the last word on performance for this. Not by a long shot.

And to drop a hint. A fairly massive hint … this is not, by any stretch of the imagination, our fastest box.

We’ll save turning things up to 11 for a later date.

I did say something about coming roaring across the 10 year mark … didn’t I. Just getting started though … just getting started 😀

[update] I guess I should point out that (in response to an inquiry)

  • Yes, they are shipping now
  • Yes, we can get pretty good densities per machine

And for those looking to build really insanely fast storage systems, as these are about 1% of 1TB/s, 100 units or so ought to be able to handle that sort of data read rate. Thats about 13 racks of these. 30 racks if you want to write 1TB/s. Assuming 10 machines per rack BTW.

And yes, they will happily be OSSes/OSTes for Lustre, FhGFS, GlusterFS, … or even a plain old NFS server.

Viewed 101272 times by 6445 viewers


One thought on “turning siFlash past 10 (GB/s that is) …

Comments are closed.