On Tue, Mar 11, 2025 at 5:51 AM Amit Kapila <amit.kapil...@gmail.com> wrote: > > On Tue, Mar 11, 2025 at 5:00 AM Masahiko Sawada <sawada.m...@gmail.com> wrote: > > > > On Sun, Mar 9, 2025 at 11:28 PM Amit Kapila <amit.kapil...@gmail.com> wrote: > > > > > > > > > Does phase 3 also use parallelism? If so, can we try to divide the > > > ring buffers among workers or at least try vacuum with an increased > > > number of ring buffers. This would be good to do for both the phases, > > > if they both use parallelism. > > > > No, only phase 1 was parallelized in this test. In parallel vacuum, > > since it uses (ring_buffer_size * parallel_degree) memory, more pages > > are loaded during phase 1, increasing cache hits during phase 3. > > > > Shouldn't we ideally try with a vacuum without parallelism with > ring_buffer_size * parallel_degree to make the comparison better?
Right. I'll share the benchmark test results with such configuration. > Also, what could be the reason for the variation in data of phase-I? > Do you restart the system after each run to ensure there is nothing in > the memory? If not, then shouldn't we try at least a few runs by > restarting the system before each run to ensure there is nothing > leftover in memory? I dropped all page caches by executing 'echo 3 > /proc/sys/vm/drop_caches' before each run and these results are the median of 3 runs. I'll investigate it further. Regards, -- Masahiko Sawada Amazon Web Services: https://aws.amazon.com
