Re: DBWR performance

From: <>
Date: Sun, 13 May 2012 21:04:38 -0700 (PDT)
Message-ID: <14549587.3006.1336968278729.JavaMail.geo-discussion-forums_at_vbbhi10>

Adaptec is in and here are some results with raid1:
  1. Write speeds:

lgwr - 130MB/sec
dbwr - 150MB/sec
direct path - 160MB/sec

Still no cigar :-) (200).

2. Lgwr looks like limited by background process running insert select and using 100% cpu, but if I do 2 sqlplus processes doing the same concurrently then lgwr speed jumps to 150MB/sec briefly and drops back to 130MB/sec. None of 3 processes is doing 100% CPU, so there is some other bottleneck. Maybe just shared memory access and waiting for semaphores.

3. Bonnie++ shows write/read 180/330 MB/sec vs 180/180 MB/sec with mdadm. This means adaptec knows little trick called reading from 2 hdd in raid 1 concurrently and mdadm does not know it.

4. If you (adaptec) are designing pc adapter and know that it is overheating then you DO NOT write everywhere in the doc that you operating temperature tops at 55C. You take big radiator or small fan or both and mount them on your PCI card, as nvidia or ati does on all their graphic cards for last gazillion years. Customer must not be forced to put extra fan to cool your card.

5. The same goes for big capacitor (replaces battery) hanging of the card on twisted wire - it is not my problem to figure out where to mount it in AT box - this has to be integrated in PCI card.


> > 6. this is a winning combo so far (1 writer, no slaves, direct io):
> >
> > db_writer_processes=1
> > filesystemio_options=directio
> > log_buffer=81920000
> >
> > Constant write speed
> >
> > lgwr - 130MB/sec
> > dbwr - 100MB/sec
> Interesting the large log buffer. Any thoughts/results on/with other
> sizes?

It does not matter for lgwr 8m or 80M - it is about the same. Received on Sun May 13 2012 - 23:04:38 CDT

Original text of this message