On 3/12/07 6:21 AM, "Simon Riggs" <[EMAIL PROTECTED]> wrote:

> So based on those thoughts, sync_scan_offset should be fixed at 16,
> rather than being variable. In addition, ss_report_loc() should only
> report its position every 16 blocks, rather than do this every time,
> which will reduce overhead of this call.

And for N concurrent scans?

I think there is actually no need to synchronize the shared buffers at all
for synchronized scans.  The OS I/O cache will do that for us and we're just
going to interfere and pessimize by trying to divide up a local buffer.

I suggest that this be proven or disproved by running this test: measure the
performance of syncscan with the non-polluting buffer change, then measure
with Jeff's patch and non-polluting with multiple scans, then measure with
your suggested changes to synchronize the buffers.

Somewhere in that progression we'll learn more about how the multi-level
buffering really works.  I think we'll get all the shared I/O cache we need.

- Luke

---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster

Reply via email to