How to interpret PD performance data

[Darking]

Just wanted to share my PD stats.

SSD:


FC:


and finally NL:




The AO has put around 8TB in FC, 5.5TB in SSD and 4TB (i set a limit) on NL
My reason was that i wanted to utilize my flash storage as best as possible, and in my head it makes more sense to tier down when capacity on SSD gets bad, than have all data in tier2 with worse responsetimes.

Oh and im using the performance AO profile

[skumflum]

Thanks for sharing!

I will try to setup AO policy to performance and see how that goes.

I am considering changing the measurement to 24 hours since IO on NL is high after 21:00. Most user activity is 08:00 to 16:00. I’m guessing that the IO must be backup or batchjobs.

This is IO for NL over 24 hours

[Darking]

Yeah that IO looks like backup to me too

Ive setup my AO to run from 20.00 and the job is allowed to run for 6 hours.

The job looks like this:

startao -btsec -46800 -etsec 7200 -maxrunh 6 -compact auto Perf_AO_CPG

So basically it runs at 20:00 from 13 hours before (7:00 in the morning), until 2 hours before (18:00).

That means it does not move blocks that are running in my backup window from 22:00 to 6:30

[afidel]

We do much the same (6:00-18:00). If backups are touching that much stale data it's probably worth reevaluating how you are doing your backups.

[skumflum]

Just a quick update… after I changed the AO policy to performance the IO pressure on the NL disk has slowly decreased.

This is the development over the last 30 days.

[Darking]

Glad that it worked for you