On Mon, Jan 26, 2015 at 11:20 PM, Robert Haas <robertmhaas@gmail.com> wrote:
> This developed a slight merge conflict. I've rebased the attached
> version, and I also took the step of getting rid of buf_table.c, as I
> think I proposed somewhere upthread. This avoids the overhead of
> constructing a BufferTag only to copy it into a BufferLookupEnt, plus
> some function calls and so forth. A quick-and-dirty test suggests
> this might not have cut down on the 1-client overhead much, but I
> think it's worth doing anyway: it's certainly saving a few cycles, and
> I don't think it's complicating anything measurably.
So here are median-of-three results for 5-minute read-only pgbench
runs at scale factor 1000, shared_buffers = 8GB, on hydra (POWER7) at
various client counts.
clients - master@4b2a25 - master+chash-buftable-v2
1 8319.254199 8105.479632
2 15485.561948 15138.227533
3 23690.186576 23264.702784
4 32157.346740 31536.870881
5 40879.532747 40455.794841
6 49063.279970 49625.681573
7 57518.374517 57492.275197
8 65351.415323 65622.211763
16 126166.416498 126668.793798
24 155727.916112 155587.414299
32 180329.012019 179543.424754
40 201384.186317 200109.614362
48 222352.265595 225688.574611
56 240400.659338 254398.144976
64 253699.031075 266624.224699
72 261198.336625 270652.578322
80 264569.862257 270332.738084
That's extremely unimpressive. You (Andres) showed a much bigger
benefit on a different machine with a much higher scale factor (5000)
so I don't know whether the modest benefit here is because of the
different machine, the different scale factor, or what.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company