[snip] I'm actually doing some very similar testing and getting very similar results. My disk is a single Seagate Barracuda 7200 RPM SATA (160 GB). The OS is openSUSE 11.1 (2.6.27 kernel) with the "stock" PostgreSQL 8.3.5 RPM. I started out running pgbench on the same machine but just moved the driver to another one trying to get better results. The other driver is quite small -- 512 MB 1.6 GHz -- so I might need to shut down the desktop and X on it. But the real mystery is this: I have two XFS partitions. Let's call them sda5 and sda6. The PostgreSQL install put the database in /var/lib/pgsql, which is on sda5. But I created a tablespace on sda6 specifically for the database that pgbench is using, and put that database there. At least that's what pgadmin3 is telling me I did. But when I run pgbench, I see intense I/O on *both* partitions sometimes, and other times I see it *only* on sda5. I can understand the "both" times -- I didn't move any "system-level" things like the write-ahead logs. But what I can't understand is the periods when it isn't using sda6, where the tablespace is. Anyhow, here's a short segment of the results I'm getting. starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 2 number of transactions per client: 100 number of transactions actually processed: 200/200 tps = 37.360964 (including connections establishing) tps = 37.430501 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 4 number of transactions per client: 100 number of transactions actually processed: 400/400 tps = 51.768918 (including connections establishing) tps = 51.985556 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 6 number of transactions per client: 100 number of transactions actually processed: 600/600 tps = 51.462103 (including connections establishing) tps = 51.734119 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 8 number of transactions per client: 100 number of transactions actually processed: 800/800 tps = 44.316328 (including connections establishing) tps = 44.473483 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 10 number of transactions per client: 100 number of transactions actually processed: 1000/1000 tps = 44.750672 (including connections establishing) tps = 44.910703 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 12 number of transactions per client: 100 number of transactions actually processed: 1200/1200 tps = 45.048743 (including connections establishing) tps = 45.205084 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 14 number of transactions per client: 100 number of transactions actually processed: 1400/1400 tps = 26.849217 (including connections establishing) tps = 26.916643 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 16 number of transactions per client: 100 number of transactions actually processed: 1600/1600 tps = 11.187072 (including connections establishing) tps = 11.198109 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 18 number of transactions per client: 100 number of transactions actually processed: 1800/1800 tps = 38.183978 (including connections establishing) tps = 38.301026 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 20 number of transactions per client: 100 number of transactions actually processed: 2000/2000 tps = 35.012091 (including connections establishing) tps = 35.109165 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 22 number of transactions per client: 100 number of transactions actually processed: 2200/2200 tps = 28.286106 (including connections establishing) tps = 28.350341 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 24 number of transactions per client: 100 number of transactions actually processed: 2400/2400 tps = 29.285593 (including connections establishing) tps = 29.358284 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 26 number of transactions per client: 100 number of transactions actually processed: 2600/2600 tps = 29.237558 (including connections establishing) tps = 29.308422 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 28 number of transactions per client: 100 number of transactions actually processed: 2800/2800 tps = 35.251509 (including connections establishing) tps = 35.351999 (excluding connections establishing) starting vacuum...end. transaction type: TPC-B (sort of) scaling factor: 100 number of clients: 30 number of transactions per client: 100 number of transactions actually processed: 3000/3000 tps = 29.790523 (including connections establishing) tps = 29.863336 (excluding connections establishing) I'm going to move the pgbench database back to the main sda5 partition -- given the "both" periods, the seeks must be killing me. This is actually not a benchmark, but a way of generating sample "blktrace" data, so when it's finally working, I'll have some block-layer traces to tell me exactly what's going on. :) -- M. Edward (Ed) Borasky I've never met a happy clam. In fact, most of them were pretty steamed. -- Sent via pgsql-performance mailing list (pgsql-performance@xxxxxxxxxxxxxx) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-performance
