Re: Benchmark-Comparison PostGreSQL vs. SQL Server

> 1. What configuration changes have you made?

None, both installations are default configured.

> 2. How many concurrent connections was this?

One.

> 3. Were you selecting 1000 rows (LIMIT 1000), selecting all the rows
> (and only fetching 1000) or actually defining an SQL cursor.

I used "LIMIT 1000", resp. "TOP 1000" statements.

> 4. What was the load on the machine - CPU or DISK peaking?
> 5. What was the RAM usage like?

CPU (< 10%) and RAM (<10%) usage were very low.

- I guess (as you mentioned below) it's the NT file system.
When running PostgreSQL queries I can hear the harddisk buzzing, but not with SQL Server queries.

On the other hand, if an application has to "fight" against the file system, I would suppose it to increase RAM and CPU usage significantly.

-----Original Message-----
From: Richard Huxton [mailto:dev(at)archonet(dot)com]
Sent: Thursday, November 25, 2004 12:34 PM
To: Robert Soeding
Cc: pgsql-general(at)postgresql(dot)org
Subject: Re: [GENERAL] Benchmark-Comparison PostGreSQL vs. SQL Server

Robert Soeding wrote:
> Hi, this is my first question here, and also, it's somewhat delicate.
> So please be patient.
>
> My question is, CAN PostGreSQL perform in the SQL Server area when it
> comes to speed? In other words, are there explanations for the results
> I found (see below)?

Faster in some cases, slower in others in my experience. Oh, and publishing performance comparisons with another database might be in breach of your ms-sql server licencing.

> Thanks, Robert
>
> ----- Background: 1. I read people were using PostGreSQL with
> TeraBytes of data sometimes, or thousands of users. These are things
> that could easily break SQL Server. - So I thought PostGreSQL might be
> similar fast to SQL Server.

Some people have very large installations. This obviously isn't on Windows, and not necessarily on x86 hardware.

> 2. I did some tests: Windows XP SP2 Several GIGs free harddisk, ~400
> MB free RAM Java 1.5 / JDBC PostGreSQL 8.0 beta (through Windows
> Installer), default configuration, default driver SQL Server 2000
> SP3a, default configuration, JDTS driver Tablespaces of both databases
> on the same partition Write-Test: Creating tables (slightly modified
> TCP-W benchmark) Read-Test: Simple SELECT statements on all tables,
> returning the first 1000 rows (cursor variants: read-only and
> non-locking, resp. updatable and locking)
>
> Results: Writing: SQL Server 25 times faster. Reading: SQL Server 100
> times faster.

The figures sound wrong. The Windows port isn't likely to be as fast as the *nix versions (certainly not yet) but those figures don't match for my experience with PG on Linux.

Unfortunately, although you provide a lot of information, almost none of it tells us what the problem is. So - 1. What configuration changes have you made?
2. How many concurrent connections was this?
3. Were you selecting 1000 rows (LIMIT 1000), selecting all the rows (and only fetching 1000) or actually defining an SQL cursor.
4. What was the load on the machine - CPU or DISK peaking?
5. What was the RAM usage like?

--
Richard Huxton
Archonet Ltd