Re: I/O on select count(*)

From: Robert Lor <Robert(dot)Lor(at)Sun(dot)COM>
To: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: Jan de Visser <jdevisser(at)digitalfairway(dot)com>, Greg Smith <gsmith(at)gregsmith(dot)com>, Alvaro Herrera <alvherre(at)commandprompt(dot)com>, pgsql-performance(at)postgresql(dot)org
Subject: Re: I/O on select count(*)
Date: 2008-05-15 21:23:10
Message-ID: 482CA9BE.8060204@sun.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance

Tom Lane wrote:
> Hmm, the problem would be trying to figure out what percentage of writes
> could be blamed solely on hint-bit updates and not any other change to
> the page. I don't think that the bufmgr currently keeps enough state to
> know that, but you could probably modify it easily enough, since callers
> distinguish MarkBufferDirty from SetBufferCommitInfoNeedsSave. Define
> another flag bit that's set only by the first, and test it during
> write-out.
>
Ok, I made a few changes to bufmgr per my understanding of your
description above and with my limited understanding of the code. Patch
is attached.

Assuming the patch is correct, the effect of writes due to hint bits is
quite significant. I collected the data below by runing pgbench in one
terminal and psql on another to run the query.

Is the data plausible?

-Robert

--------------

Backend PID : 16189
SQL Statement : select count(*) from accounts;
Execution time : 17.33 sec

============ Buffer Read Counts ============
Tablespace Database Table Count
1663 16384 2600 1
1663 16384 2601 1
1663 16384 2615 1
1663 16384 1255 2
1663 16384 2602 2
1663 16384 2603 2
1663 16384 2616 2
1663 16384 2650 2
1663 16384 2678 2
1663 16384 1247 3
1663 16384 1249 3
1663 16384 2610 3
1663 16384 2655 3
1663 16384 2679 3
1663 16384 2684 3
1663 16384 2687 3
1663 16384 2690 3
1663 16384 2691 3
1663 16384 2703 4
1663 16384 1259 5
1663 16384 2653 5
1663 16384 2662 5
1663 16384 2663 5
1663 16384 2659 7
1663 16384 16397 8390

======== Dirty Buffer Write Counts =========
Tablespace Database Table Count
1663 16384 16402 2
1663 16384 16394 11
1663 16384 16397 4771

========== Hint Bits Write Counts ==========
Tablespace Database Table Count
1663 16384 16397 4508

Total buffer cache hits : 732
Total buffer cache misses : 7731
Average read time from cache : 9136 (ns)
Average read time from disk : 384201 (ns)
Average write time to disk : 210709 (ns)

Backend PID : 16189
SQL Statement : select count(*) from accounts;
Execution time : 12.72 sec

============ Buffer Read Counts ============
Tablespace Database Table Count
1663 16384 16397 8392

======== Dirty Buffer Write Counts =========
Tablespace Database Table Count
1663 16384 16394 6
1663 16384 16402 7
1663 16384 16397 2870

========== Hint Bits Write Counts ==========
Tablespace Database Table Count
1663 16384 16402 2
1663 16384 16397 2010

Total buffer cache hits : 606
Total buffer cache misses : 7786
Average read time from cache : 6949 (ns)
Average read time from disk : 706288 (ns)
Average write time to disk : 90426 (ns)

Attachment Content-Type Size
hint-bits-test.patch text/x-patch 4.6 KB

In response to

Browse pgsql-performance by date

  From Date Subject
Next Message James Mansion 2008-05-15 22:11:52 Re: I/O on select count(*)
Previous Message Subbiah Stalin-XCGF84 2008-05-15 19:27:48 Re: Update performance degrades over time