"Douglas McNaught" <doug@mcnaught.org> writes:
> On 12/2/07, Gregory Stark <stark@enterprisedb.com> wrote:
>>
>> The two interfaces I'm aware of for this are posix_fadvise() and libaio. I've
>> run tests with a synthetic benchmark which generates a large file then reads a
>> random selection of blocks from within it using either synchronous reads like
>> we do now or either of those interfaces. I saw impressive speed gains on a
>> machine with only three drives in a raid array. I did this a while ago so I
>> don't have the results handy. I'll rerun the tests again and post them.
>
> The issue I've always seen raised with asynchronous I/O is
> portability--apparently some platforms PG runs on don't support it (or
> not well). AIUI Linux actually still has a fairly crappy
> implementation of AIO--glibc starts threads to do the I/O and then
> tracks when they finish. Not absolutely horrible, but a nice way to
> suddenly have a threaded backend when you're not expecting one.
In the tests I ran Linux's posix_fadvise worked well and that's the simpler
interface for us to adapt to anyways. On Solaris there was no posix_fadvise
but libaio worked instead.
-- Gregory Stark EnterpriseDB http://www.enterprisedb.com Get trained by Bruce Momjian - ask me about
EnterpriseDB'sPostgreSQL training!
