SAN disk performance requzirement

Support
1

Problem:
Foreman-maintain healt-check --label disk performance
Interesting issue in lot of systems - attached to SAN - I wwill get warning we land below 60MB/s.

As far I see, the health check woul like to test raw 1 stream, direct IO performance. It is fine with local HDD’s, but issue over SAN access.

What should be the correct output or the correct fio command for network ( even iscsi or FC) attached storage? Has anybody any good measurement command for?

Actually the health check uses:
fio --name=job1 --rw=read --size=1g --directory=#directory --direct=1 --unlink=1

sample values: I get 10MB/s but with - direct=0 1500MB/s ( ok good cache) - or e.g. witrh test using libaio with direct=1 - from same system 250MB/s.

Really need we that raw sync perfromance?

Thanks
hkicsi

2

Dear All,

I would like to extend my observations, probably somebody has done already some activity. The most annoying thing postgres checkpointer runs constantly: I think - and correct me pls if I’m wrong: postgres is one of our main storage performance consumer - for random block IO.
Now - I have found in enterprise environments:
"psql -d foreman -c “select * from pg_stat_bgwriter;”
shows buffer_clean lower as buffer_backend - probably even 0, and maxwrites clean not 0. On my test system ( with local ssd’s) I never observed such postgres behaviour.
Can it be a sign about lack of performance from stoage side? and much more important what about postgres tuning? Has anybody expereience about?

Thanks
hkicsi