Dne 26.5.2011 16:39, Merlin Moncure napsal(a):
> On Thu, May 26, 2011 at 9:01 AM, Tomas Vondra <tv@fuzzy.cz> wrote:
>> Dne 26.5.2011 11:41, Alexander Farber napsal(a):
>>> Also I wonder, how's shared memory used by PostgreSQL.
>>> I'm irritated - how it could work with 32MB,
>>> but now I've got suggestion to increase it
>>> to 512MB (and it seems to work too...)
>>
>> Shared buffers are a 'database cache'. When the DB needs a block from a
>> file (because that's where the data are stored), it reads the data into
>> the cache. When the same block is needed again, it may be read from the
>> cache (which is much faster). Unless there's not enough space to hold
>> all the blocks - in that case the block may be removed from the cache
>> and will be read from the disk again.
>
> *or the disk cache*. lowering shared buffers does not lower the
> amount of ram in the system and thus does not lower the availability
> of cache. If I may nitpick this point on your otherwise very
> excellent email, this is exactly the type of thing that drives me
> crazy about advice to raise shared buffers. It suggests you will get
> less disk i/o which may or may not be the case (in fact, it can make
> the i/o problem worse). If it does help i/o, it will probably not be
> for the reasons you suspect. See my thread in -performance on this
> topic.
Yes, you're right. I didn't want to complicate the things further so
I've skipped the part about page cache.
Tomas