Actually, the reason for utilizing all available RAM is to improve=20=20
efficiency. There is no cost to leaving things in core, in case they=20=20
will can be re-used. It takes virtually no time to make pages=20=20
available when they are needed. The only cost is in managing=20=20
utilization maps, but accessing the hard drive takes milliseconds=20=20
where getting something from RAM takes nanoseconds. It's really an=20=20
operating system thing, not a Postgres issue.
On Oct 17, 2007, at 9:04 PM, ssurui wrote:
> Thank you for the explanations.
>
> My program is running 4 months, it has never exit because of the=20=20
> shortage of memory. But the =91free=92 memory (observing by top=20=20
> command) is kept on about 20MB (The database server has 2GB DDR RAM).
>
>
>
> According to you explanations, when I =93malloc=94 a buffer of 30MB,=20=
=20
> the cache will be reduced to give me 30MB. I think it is true. So=20=20
> another question is whether the efficiency of program is affected=20=20
> to go down?
>
>=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=
=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=20=
=20=20=20=20
> ssurui@mailst.xjtu.edu.cn
>
>
>
> ----------------------------------------------------------------------=20
> ----------------------------------------------------------------------=20
> --
>
> I think the one thing everyone knows as a fact is that the memory=20=20
> did not vanish. Your latest message gives us a clue that what=20=20
> interests you is why free (unused, idle, not giving you any=20=20
> benefit) memory went down. If that is the question, it is because=20=20
> PostgreSQL goes through the OS file systems, so RAM which is=20=20
> serving no other purpose will automatically be used to cache data.=20=20=
=20
> If you need memory for something else, the cache will be reduced to=20=20
> compensate.
>
>
>
> -Kevin
>
>
Douglas Toltzman
doug@oakstreetsoftware.com
(910) 526-5938
