[lug] Debugging I/O performance

[Rob Nagler]

It's the SSDs. I swapped various bits.  See https://github.com/radiasoft/d
evops/issues/142 for doc.

Some things I learned, some of which were stated already (thanks
everybody!):

* Dell H310 is 20% faster than mobo AHCI
* Identical model/firmware SSD speeds vary widely
* Smaller SSDs are faster than larger
* SMART stats don't correlate to performance (see)

I also tested good old spinning disks (WD 500g Scorpio Black), and even the
slowest SSDs were more than 2x faster, and the fastest SSDs were 5x. The
spinning disks were extremely consistent even though they were different
versions bought at different times and used differently. Slow and stead
doesn't win the race. :)

I need to look at the SMART stats on the SSDs, and there doesn't seem to be
a great correlation. One of f3's drives was more used than both of f2's,
but the performance is much worse for f2 (all four drives are identical
firmware). f4 has a drive which has 6459E6 writes, but it's drives are as
fast as f3's, and much faster than f7's and f2's. All of these are RAID1
(slowest drive wins).

Maybe others see a pattern in the data. I'm not sure what to conclude
except that you have to monitor speed (not stats) to determine how well a
disk is performing.

Rob
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lug.boulder.co.us/pipermail/lug/attachments/20180325/852e383c/attachment.html>