> It wouldn't be a very bad idea for systems where mmap is
> noticeably faster than read/write using syscalls.
> Unfortunately on some of those systems mmap is broken for
> multiple processes mapping the same file...:)
Yeah, fair enough. But mmap works well on the more popular platforms
used for PostgreSQL. And it can't *hurt* performance, and its probably
worth doing simply so that PostgreSQL "plays nicely" with other
applications using the VM resources on a particular system, instead of
the "fixed size buffer cache" approach.
> But if someone wants to work on it, this would be fairly
> modest-sized project that only affects bufmgr...
Interesting.. might be worth taking a look at..
- Andrew
---------------------------(end of broadcast)---------------------------
TIP 2: you can get off all lists at once with the unregister command
(send "unregister YourEmailAddressHere" to [EMAIL PROTECTED])