Hi, I've got some interesting (IMHO) results of comparison between two
servers which may be gilt-edged.
(Simply: Postgres 7.4.3 with fsync turned off, Tomcat 5.0.25 &
j2sdk-1.4.2-rc1).
Opteron (SID, x86-64; kernel: 2.6.7):
MB: Tyan Thunder K8S S2880
2x Opteron 246 (2,0Ghz) 1Mb
4Gb Ram ECC Reg. PC3200
RAID ADAPTEC 2120S
3x Seagate SCSI 10000rpm 73Gb->RAID5
Xeon (SID, i386; kernel: 2.6.4-p4-aic79) :
MB: INTEL SE7501WV2 SCSI
2x Xeon 3,06 (HT ON) 512KB
2Gb Ram ECC Reg. PC2100
Integrated SCSI Adaptec AIC-7902
2x Seagate SCSI 15000rpm 18Gb ST318453LC ->soft raid
CATALINA_OPTS="-Xmx600m -server"
====================
ab -n 20000 -c 500
http://localhost:8080/jsp-examples/jsp2/simpletag/repeat.jsp
results:
Opteron:
Requests per second: 5878.89 [#/sec] (mean)
Time per request: 85.05 [ms] (mean)
Time per request: 0.17 [ms] (mean, across all concurrent requests)
Transfer rate: 10670.73 [Kbytes/sec] received
Xeon:
Requests per second: 2212.88 [#/sec] (mean)
Time per request: 225.95 [ms] (mean)
Time per request: 0.45 [ms] (mean, across all concurrent requests)
Transfer rate: 4018.18 [Kbytes/sec] received
====================
pgbench with read/write transactions
pgbench -c 10 -t 100 pgbench
Opteron:
tps = 1011.802678 (including connections establishing)
tps = 1074.345804 (excluding connections establishing)
Xeon:
tps = 514.228179 (including connections establishing)
tps = 531.858596 (excluding connections establishing)
====================
pgbench with read only transactions
pgbench -c 10 -t 100 -S pgbench
Opteron:
tps = 2787.650707 (including connections establishing)
tps = 3309.417278 (excluding connections establishing)
Xeon:
tps = 2508.132620 (including connections establishing)
tps = 2987.241492 (excluding connections establishing)
=======================================
Summary:
1. simple http query test - opteron two times faster
2. full TPC-B-like pgbench test - opteron two times faster
3. read only pbgench test - opteron ca. 10% faster
