Tom Lane wrote:
Nope, Pavan's nailed it: the problem is that after using a buffer, the
seqscan leaves it with usage_count = 1, which means it has to be passed
over once by the clock sweep before it can be re-used. I was misled in
the 32-buffer case because catalog accesses during startup had left the
buffer state pretty confused, so that there was no long stretch before
hitting something available. With a large number of buffers, the
behavior is that the seqscan fills all of shared memory with buffers
having usage_count 1. Once the clock sweep returns to the first of
these buffers, it will have to pass over all of them, reducing all of
their counts to 0, before it returns to the first one and finds it now
usable. Subsequent tries find a buffer immediately, of course, until we
have again filled shared_buffers with usage_count 1 everywhere. So the
problem is not so much the clock sweep overhead as that it's paid in a
very nonuniform fashion: with N buffers you pay O(N) once every N reads
and O(1) the rest of the time. This is no doubt slowing things down
enough to delay that one read, instead of leaving it nicely I/O bound
all the time. Mark, can you detect "hiccups" in the read rate using
your setup?
Cool. You posted the same analysis before I could hit the "send" button :)
I am wondering whether seqscan would set the usage_count to 1 or to a higher
value. usage_count is incremented while unpinning the buffer. Even if
we use
page-at-a-time mode, won't the buffer itself would get pinned/unpinned
every time seqscan returns a tuple ? If thats the case, the overhead would
be O(BM_MAX_USAGE_COUNT * N) for every N reads.
I seem to recall that we've previously discussed the idea of letting the
clock sweep decrement the usage_count before testing for 0, so that a
buffer could be reused on the first sweep after it was initially used,
but that we rejected it as being a bad idea. But at least with large
shared_buffers it doesn't sound like such a bad idea.
How about smaller value for BM_MAX_USAGE_COUNT ?
Another issue nearby to this is whether to avoid selecting buffers that
are dirty --- IIRC someone brought that up again recently. Maybe
predecrement for clean buffers, postdecrement for dirty ones would be a
cute compromise.
Can we use a 2-bit counter where the higher bit is set if the buffer is
dirty
and lower bit is set whenever the buffer is used. The clock-sweep then
decrement this counter and chooses a victim with counter value 0.
ISTM that we should optimize for large shared buffer pool case,
because that would be more common in the coming days. RAM is
getting cheaper everyday.
Thanks,
Pavan
---------------------------(end of broadcast)---------------------------
TIP 2: Don't 'kill -9' the postmaster