Hi,
I have a machine with one SATA SSD and one NVMe device. Of course
even the SSD is pretty high performance and the NVMe is over 4 times
faster than that, if we're talking about IOPS.
I've never used --write-mostly before but I thought maybe if I put
them both in a RAID-10 but set the SSD device to --write-mostly then
most reads would come from the NVMe and benefit from its increased
performance. I have not, however, been able to demonstrate that. It
doesn't appear to do anything different compared to not using
--write-mostly.
Here's some fio stats:
https://tools.bitfolk.com/wiki/User:Strugglers/write-mostly
Am I testing it incorrectly or are my expectations wrong? Is it just
not expected to do anything with devices like these?
I realised that I only tested things with a single fio process with
a queue depth of 32, so I tried some more tests with 4 processes and
a queue depth of 64, but again the results were indistinguishable.
Debian testing (buster), which has mdadm package version 4.1-1 and
kernel package version linux-image-4.19.0-4-amd64 4.19.28-2. The git
HEAD version of fio was used.
Cheers,
Andy
