Thread: Default shared memory SGML mention
I am looking over the shared memory SGML files, and found that on my
BSD/OS system, I get as the following default shared memory size:
m 131072 5432001 --rw------- postgres user postgres
user 1 1236992 10622 10622 12:11:14 12:11:15 12:11:14
^^^^^^^
I have highlighted the size. If I subtract the default shared memory
buffers, I get:
1236992 - 8192 * 64
712704
so by my computations, we need about 700k added to the shared buffer
cache area. The documentation says to add 512k, not 700k. My guess is
that the non-cached shared memory area has increased in a recent
release, and docs were not updated. I will bump up the SGML mention to
700k unless someone shows a lower default shared memory size.
--
Bruce Momjian | http://candle.pha.pa.us
pgman@candle.pha.pa.us | (610) 853-3000
+ If your life is a hard drive, | 830 Blythe Avenue
+ Christ can be your backup. | Drexel Hill, Pennsylvania 19026
Bruce Momjian <pgman@candle.pha.pa.us> writes:
> so by my computations, we need about 700k added to the shared buffer
> cache area. The documentation says to add 512k, not 700k.
This would depend on your -N setting.
A little experimentation shows that the present multiplier for -B is
about 8350 bytes and the multiplier for -N about 14580 bytes on my
system (HPUX), which extrapolates to a fixed base of about 250K plus
the -B and -N contributions. In reality I believe the computation
is somewhat nonlinear, but that's probably close enough for starters.
regards, tom lane
> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > so by my computations, we need about 700k added to the shared buffer
> > cache area. The documentation says to add 512k, not 700k.
>
> This would depend on your -N setting.
>
> A little experimentation shows that the present multiplier for -B is
> about 8350 bytes and the multiplier for -N about 14580 bytes on my
> system (HPUX), which extrapolates to a fixed base of about 250K plus
> the -B and -N contributions. In reality I believe the computation
> is somewhat nonlinear, but that's probably close enough for starters.
I have updated the SGML sources to say for default shared memory values:
250 kB + 8192 * buffers or infinity
Of course, buffers is 2 * connections by default.
--
Bruce Momjian | http://candle.pha.pa.us
pgman@candle.pha.pa.us | (610) 853-3000
+ If your life is a hard drive, | 830 Blythe Avenue
+ Christ can be your backup. | Drexel Hill, Pennsylvania 19026
> Bruce Momjian <pgman@candle.pha.pa.us> writes: > >> A little experimentation shows that the present multiplier for -B is > >> about 8350 bytes and the multiplier for -N about 14580 bytes on my > >> system (HPUX), which extrapolates to a fixed base of about 250K plus > >> the -B and -N contributions. In reality I believe the computation > >> is somewhat nonlinear, but that's probably close enough for starters. > > > I have updated the SGML sources to say for default shared memory values: > > > 250 kB + 8192 * buffers or infinity > > How did infinity get into this? And what happened to the -N > contribution? Infinity was always there. The column is labeled "Typical Values". As far as -N, buffers is always 2*N, right, so I figured it was in there already. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
Bruce Momjian <pgman@candle.pha.pa.us> writes:
>> A little experimentation shows that the present multiplier for -B is
>> about 8350 bytes and the multiplier for -N about 14580 bytes on my
>> system (HPUX), which extrapolates to a fixed base of about 250K plus
>> the -B and -N contributions. In reality I believe the computation
>> is somewhat nonlinear, but that's probably close enough for starters.
> I have updated the SGML sources to say for default shared memory values:
> 250 kB + 8192 * buffers or infinity
How did infinity get into this? And what happened to the -N
contribution?
regards, tom lane
Bruce Momjian <pgman@candle.pha.pa.us> writes:
> far as -N, buffers is always 2*N, right, so I figured it was in there
> already.
No, it isn't, and buffers is NOT always 2*N.
regards, tom lane
> Bruce Momjian <pgman@candle.pha.pa.us> writes: > > far as -N, buffers is always 2*N, right, so I figured it was in there > > already. > > No, it isn't, and buffers is NOT always 2*N. Yes, but shared memory limits are based on buffers, not connections, right? The per-connection overhead is pretty small, no? If it is buffers and connections, how much is connections and how much buffers. I know buffers is 8k. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Yes, but shared memory limits are based on buffers, not connections,
> right? The per-connection overhead is pretty small, no?
I gave you my experimentally-derived formula already:
250Kb + 8350 * B + 14580 * N
I don't see how you can consider the N term negligible.
regards, tom lane
> Bruce Momjian <pgman@candle.pha.pa.us> writes: > > Yes, but shared memory limits are based on buffers, not connections, > > right? The per-connection overhead is pretty small, no? > > I gave you my experimentally-derived formula already: > > 250Kb + 8350 * B + 14580 * N > > I don't see how you can consider the N term negligible. Oh, I misread it. I thought the 14.5k was part of the 250k. Let me fix that. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026
> Bruce Momjian <pgman@candle.pha.pa.us> writes: > > Yes, but shared memory limits are based on buffers, not connections, > > right? The per-connection overhead is pretty small, no? > > I gave you my experimentally-derived formula already: > > 250Kb + 8350 * B + 14580 * N > > I don't see how you can consider the N term negligible. OK, patch applied. -- Bruce Momjian | http://candle.pha.pa.us pgman@candle.pha.pa.us | (610) 853-3000 + If your life is a hard drive, | 830 Blythe Avenue + Christ can be your backup. | Drexel Hill, Pennsylvania 19026 Index: doc/src/sgml/runtime.sgml =================================================================== RCS file: /home/projects/pgsql/cvsroot/pgsql/doc/src/sgml/runtime.sgml,v retrieving revision 1.64 diff -c -r1.64 runtime.sgml *** doc/src/sgml/runtime.sgml 2001/05/07 15:55:27 1.64 --- doc/src/sgml/runtime.sgml 2001/05/08 21:09:21 *************** *** 1481,1487 **** <row> <entry><varname>SHMMAX</></> <entry>Maximum size of shared memory segment (bytes)</> ! <entry>250 kB + 8192 * buffers or infinity</entry> </row> <row> --- 1481,1487 ---- <row> <entry><varname>SHMMAX</></> <entry>Maximum size of shared memory segment (bytes)</> ! <entry>250 kB + 8.2kB * buffers + 14.2kB * max_connections or infinity</entry> </row> <row>