Hello Tom,
Tom Lane wrote:
So on the strength of that, I'm going to go ahead and commit the patch,
but I'd be interested to see benchmarks from people with access to
better hardware.
I've just completed two dbt2 test runs on a mid-level system, with 4GB
RAM and a 7 disk SATA RAID 1+0 w/ BBU. Once with code as of 2007/09/05
18:00 (which is *before* the first lazy xid commit) and once with cvs
HEAD (2007/09/08 +latestCompletedXid.patch.
Here are the results from the first test run (test run 33, without lazy
xid):
$ cat 33/driver/results.out
Response Time (s)
Transaction % Average : 90th % Total Rollbacks %
------------ ----- --------------------- ----------- --------------- -----
Delivery 3.97 3.745 : 7.844 11844 0 0.00
New Order 45.35 3.844 : 7.692 135192 1352 1.01
Order Status 3.95 2.728 : 6.371 11764 0 0.00
Payment 42.74 2.649 : 6.349 127415 0 0.00
Stock Level 4.00 2.172 : 5.634 11915 0 0.00
------------ ----- --------------------- ----------- --------------- -----
1103.45 new-order transactions per minute (NOTPM)
120.1 minute duration
0 total unknown errors
1003 second(s) ramping up
And that's with HEAD +latestCompletedXid.patch (test run 34):
$ cat 34/driver/results.out
Response Time (s)
Transaction % Average : 90th % Total Rollbacks %
------------ ----- --------------------- ----------- --------------- -----
Delivery 3.96 3.843 : 8.223 11760 0 0.00
New Order 45.28 4.049 : 8.451 134398 1300 0.98
Order Status 3.97 2.877 : 6.815 11777 0 0.00
Payment 42.80 2.745 : 6.718 127027 0 0.00
Stock Level 4.00 2.280 : 6.129 11859 0 0.00
------------ ----- --------------------- ----------- --------------- -----
1097.71 new-order transactions per minute (NOTPM)
120.1 minute duration
0 total unknown errors
1003 second(s) ramping up
Both tests ran for two hours, had 100 warehouses and 50 connections.
shared_buffers were set to 1024MB, effective_cachesize = 3800MB, all
other settings were standard.
Regards
Markus
---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
choose an index scan if your joining column's datatypes do not
match