Re: FSM search modes

From: "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov>
To: "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us>
Cc: "Simon Riggs" <simon(at)2ndQuadrant(dot)com>, "Alvaro Herrera" <alvherre(at)commandprompt(dot)com>, "decibel" <decibel(at)decibel(dot)org>, "Heikki Linnakangas" <heikki(dot)linnakangas(at)enterprisedb(dot)com>, "Itagaki Takahiro" <itagaki(dot)takahiro(at)oss(dot)ntt(dot)co(dot)jp>, "pgsql-hackers" <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: FSM search modes
Date: 2009-10-02 14:03:01
Message-ID: 4AC5C1C5020000250002B5BC@gw.wicourts.gov
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov> writes:
>> [pages with free space or total pages in relation?]
>
> It's going to be the latter --- we do not know, and are *not* going
> to invest the cycles to find out, how many pages have a useful
> amount of free space. Even finding out the relation physical length
> might be more cycles than we want to spend here ...

OK, after mulling this over for a while, I suspect we'd do pretty well
with starting to consider resetting the sweep after hitting 50% of the
last known relation size (or whatever best approximation is available
at low cost), and using a reset probability of 1 / (max_connections *
4). That gives about a 77.8% chance of getting to at least
max_connections before resetting the sweep. Since it leaves about an
8.2% chance of getting to at least ten times max_connections pages
before resetting the sweep, I'm inclined to think we'd want to start
that at 50% of the relation rather than waiting until we get to the
last quarter. As one more data point to consider, if someone has 2000
connections configured (which I've seen mentioned in many posts), you
would get to 50,000 pages past the point where you start using this
technique one time out of 500.

-Kevin

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message Pavel Stehule 2009-10-02 14:06:45 Re: Issues for named/mixed function notation patch
Previous Message Michael Meskes 2009-10-02 13:37:58 Re: CommitFest 2009-09, two weeks on