Re: FW: Tx forecast improving harware capabilities.

Поиск
Список
Период
Сортировка
От Josh Berkus
Тема Re: FW: Tx forecast improving harware capabilities.
Дата
Msg-id 200508180930.48089.josh@agliodbs.com
обсуждение исходный текст
Ответ на FW: Tx forecast improving harware capabilities.  ("Sebastian Lallana" <slallana@datatransfer.com.ar>)
Список pgsql-performance
Sebastian,

> We are having serious performance problems using JBOSS and PGSQL.

How about some information about your application?   Performance tuning
approaches vary widely according to what you're doing with the database.

Also, read this:
http://www.powerpostgresql.com/PerfList

> I'm sure the problem has to do with the application itself (and neither
> with JBOSS nor PGSQL) but the fact is that we are using desktop
> equipment to run both Jboss and Postgresql (An Athlon 2600, 1 Gb Ram,
> IDE HDD with 60 Mb/sec Transfer Rate), and the answers arise:

Well, first off, the IDE HDD is probably killing performance unless your
application is 95% read or greater.

> If we upgrade our hardware to a Dual Processor would the transactions
> per second increase significantly? Would Postgresql take advantage from
> SMP? Presumably yes, but can we do a forecast about the number of tps?

If this is an OLTP application, chance are that nothing is going to improve
performance until you get decent disk support.

> What we need is a paper with some figures showing the expected
> performance in different environments. Some study about the "degree of
> correlation" between TPS and Number of Processors, Cache, Frequency,
> Word Size, Architecture, etc.

I don't think such a thing exists even for Oracle.   Hardware configuration
for maximum performance is almost entirely dependant on your application.

If it helps, running DBT2 (an OLTP test devised by OSDL after TPC-C), I can
easily get 1700 new orders per minute (NOTPM) (about 3000 total
multiple-write transactions per minute) on a quad-pentium-III with 4GB RAM
and 14 drives, and 6500 notpm on a dual-Itanium machine.

> P.S. I've been looking at www.tpc.org but I could't find anything
> valuable.

Nor would you for any real-world situation even if we had a TPC benchmark
(which are involved and expensive, give us a couple of years).  The TPC
benchmarks are more of a litmus test that your database system & platform are
"competitive"; they don't really relate to real-world performance (unless you
have budget for an 112-disk system!)

--
Josh Berkus
Aglio Database Solutions
San Francisco

В списке pgsql-performance по дате отправления:

Предыдущее
От: "Sebastian Lallana"
Дата:
Сообщение: FW: Tx forecast improving harware capabilities.
Следующее
От: Jeremiah Jahn
Дата:
Сообщение: Re: extremly low memory usage