Re: 8xIntel S3500 SSD in RAID10 on Dell H710p - Mailing list pgsql-performance

From Strahinja Kustudić
Subject Re: 8xIntel S3500 SSD in RAID10 on Dell H710p
Date
Msg-id CADKbJJXL+h8uxwgQF45VFtgei70v6WowReCqLh1aeS4MEAph_g@mail.gmail.com
Whole thread Raw
In response to Re: 8xIntel S3500 SSD in RAID10 on Dell H710p  ("Graeme B. Bell" <grb@skogoglandskap.no>)
List pgsql-performance
- disk cache settings (EnDskCache - for SSD should be on or you're going to lose 90% of your performance)

Disk cache is enabled, I know there is a huge performance impact.
 
- OS settings e.g.

echo noop > /sys/block/sda/queue/scheduler
echo 975 > /sys/block/sda/queue/nr_requests
blockdev --setra 16384 /dev/sdb

I'll try to play with these as well. I haven't tried noop yet, but it was on my checklist. 

 
- OS kernel version

We use H710Ps with SSDs as well, and these settings make a measurable difference to our performance here (though we measure more than just pgbench since it's a poor proxy for our use cases).

Also

- SSDs - is the filesystem aligned and block size chosen correctly (you don't want to be forced to read 2 blocks of SSD to get every data block)? RAID stripe size? May make a small difference.

I read a lot about filesystem alignment, but as far as I understand if I just format the whole drive and not created any partitions, all should be aligned. I'll check what is the data block size of SSDs and try set block size to that, but I'm not exactly sure how does block size work with RAID10, it seem logical to me that it works differently.
 

- are the SSDs all sitting on different SATA channels? You don't want them to be forced to share one channel's worth of bandwidth. The H710P has 8 SATA channels I think (?) and you mention 10 devices above.

Good question. Since the server is not physical at my place, I will have to check with the people who assembled it.


Thanks for your help. Once I do more testing, I'll return more details what helped.

pgsql-performance by date:

Previous
From: Evgeniy Shishkin
Date:
Subject: Re: Tuning the configuration
Next
From: Simon Riggs
Date:
Subject: Re: Yet another abort-early plan disaster on 9.3