Обсуждение: Question of performance of version 8

От:
kondo_yo@itg.hitachi.co.jp
Дата:

Hello.
I tried performance test of version 8.0.0 beta4 by osdl-dbt-1.
The result is that throughput of version 8 fell to about 70 percent
compared with V7.4.6.
Test result is below. (measurement was repeated 3 times)

-------------------------------------------------------
Hardware spec
 CPU   Pentium Ⅲ  531.986 MHz
 Memory  125660 Kb
-------------------------------------------------------
Software version
 OS Linux Kernel:2.4.21-4.EL
 Distribution: Red Hat Enterprise Linux AS release 3
-------------------------------------------------------
Result(transaction per second)
             first  second third
  7.4.6      6.8    6.8    5.9
  8.0 beta4   4.3    4.6   4.4
-------------------------------------------------------
Parameter of DBT1,PostgreSQL
 Database Size (items): 1000
 Database Size (customers): 10
 number of cache: 1
 number of connections between cache and database: 10
 number of application server: 1
 number of connections between server and database: 20
 number of drivers: 1
 eus/driver: 100
 rampuprate/driver: 60
 duration/driver: 900
 thinktime/driver: 1.6
 Put WAL on different driver: 0
 Put pgsql_tmp on different driver: 0
 database parameters: -i -c listen_addresses='*'
 shmmax: 33554432
-------------------------------------------------------

- Both version 8 and version 7 are performed under the same condition.
- Tuning of adjustment of a parameter was not carried out at all.
- The server and client process are performed in the same machine.

Is there the weak point on the performance in version 8.0.0 ?

Any help would greatly appreciated.

  kondou

От:
Tom Lane
Дата:

 writes:
> I tried performance test of version 8.0.0 beta4 by osdl-dbt-1.
> The result is that throughput of version 8 fell to about 70 percent
> compared with V7.4.6.

beta4 is a little bit back ...

I don't have dbt1 at hand, but I tried pg_bench on PG 7.4.6 against 8.0rc2
just now.  For the test case

    pgbench -i -s 10 bench
    pgbench -c 10 -t 10000 bench

on 7.4.6 I get:

starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 10
number of clients: 10
number of transactions per client: 10000
number of transactions actually processed: 100000/100000
tps = 65.714249 (including connections establishing)
tps = 65.716363 (excluding connections establishing)

on 8.0rc2 I get:

starting vacuum...end.
transaction type: TPC-B (sort of)
scaling factor: 10
number of clients: 10
number of transactions per client: 10000
number of transactions actually processed: 100000/100000
tps = 107.379742 (including connections establishing)
tps = 107.385301 (excluding connections establishing)

This is on a Fedora Core 3 machine, generic three-year-old PC with cheap
junk IDE drive that lies about write completion (but I'm running fsync
off so that hardly matters ;-)).  And I didn't change any of the default
postgresql.conf settings, just started the postmaster with -F in both
cases.  So I wouldn't claim that it's very representative of real-world
performance on real-world server hardware.  But I'm not seeing a serious
falloff from 7.4 to 8.0 here --- more the other way 'round.

Please try dbt1 with 8.0rc2, and let us know if you still see a problem.

            regards, tom lane

От:
kondo_yo@itg.hitachi.co.jp
Дата:

Thank you for your attention.
I will try again with new postgres release and
 examine access method of sql with explain command.
  kondo