logo-micron

Add Bookmark(s)


To:

Email


Bookmark(s) shared successfully!

Please provide at least one email address.

Microchip background

Experiences in the Home Lab With Micron P420m PCIe SSD, Pt. 3

Experiences in the Home Lab With Micron P420m PCIe SSD, Part 3

Greetings once again from the homelab!  In my last  post, I described my findings running 4KB I/Os using the Micron P420m PCIe SSD. Now, I’d like to summarize what I’ve found with 32KB I/Os – a much more realistic block size, and do a bit of comparison as well.

My lab setup was outlined in part 1 of this blog series and further described in part 2. The first results I’ll show are findings running a synthetic 32KB random write workload across a 600GB surface at various I/Os outstanding (meaning the device responds to each I/O before the host issues a subsequent request). Note, during these tests, the adapter queue depth was 255 and the device was fully preconditioned (24 hours’ worth) to insure steady state behavior.   

As in previous tests, I issued the following command to set the device queue depth to 255:

esxcli storage core device set –m 255 –O 255 –d <P420m device name>

 

Outstanding I/Os

IOPS

Average Latency (µs)

1

8445

117

2

9077

219

4

8697

459

8

8344

958

16

8019

1994

32

7580

4220

For 32KB, the "sweet spot" is at queue depth 2. Again, as in previous tests, you can see the effect of parallelism on the P420m. Still, at 1 I/O outstanding, the IOPS are more than 90% of maximum and latency, as expected, is roughly half.  117 microseconds is an outstanding figure for a 32KB random write request, one that many current databases perform. Moral of the story: Take advantage of the low latency provided by the P420m for database workloads performing ACID (atomic, consistent, isolated, durable) functions.

As mentioned in my previous blog, understanding latency distribution is important in order to judge the variability of response times. As such, here are the results from the testing:

At device queue depth 255 and 2 I/Os outstanding, there were 8,170,195 I/Os issued, of which 8,131,265 (99.5235%) completed in 1ms or less. The figure rises to 99.9947% for I/Os completed in 2ms or less.  Clearly, the P420m is at its best when parallelism is in play.

Latency Distribution Chart

In my next post, I’ll cover application testing—an even more realistic scenario. In the meantime, let us know what you think. Send us a tweet via the @MicronStorage handle, or contact me directly at @peglarr.

Login or Sign Up Now for an account to leave a comment.