Here is my result without -C pgbench -h 127.0.0.1 -p 9999 -U postgres -c 100 -t 10 -s 10 pgbench
Scale option ignored, using pgbench_branches table count = 10 starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 10 query mode: simple number of clients: 100 number of threads: 1 number of transactions per client: 10 number of transactions actually processed: 1000/1000 tps = 98.353544 (including connections establishing) tps = 196.318788 (excluding connections establishing) On Sun, Dec 19, 2010 at 1:51 AM, Jeff Janes <jeff.ja...@gmail.com> wrote: > On Sat, Dec 18, 2010 at 10:15 AM, tuanhoanganh <hatua...@gmail.com> wrote: > > I have server computer install Windows 2008R2, PostgreSQL 9.0.1 64 bit, > 8G > ... > > pgbench -h 127.0.0.1 -p 5433 -U postgres -c 100 -t 10 -C -s 10 pgbench > > Why the -C option? You are essentially benchmarking how fast you can > make new connections to the database. Is that what you want to be > benchmarking? > > If the code you anticipate using is really going to make and break > connections between every query, you should use a connection pooler. > Which means you should be benchmarking through the connection pooler, > or just leave off the -C. > > Also, -t 10 is probably too small to get meaningful results. > > > tps = 20.143494 (including connections establishing) > > tps = 256.630260 (excluding connections establishing) > > > > Why pgbench on my server is very low or is it common value with my server > ? > > Starting a new connection in PG is relatively slow, especially so on > Windows, because it involves starting and setting up a new process for > each one. > > Cheers, > > Jeff >
