Re: 2.1.115 is not slower than 2.0.34

Eric Lee Green (eric@linux-hw.com)
Mon, 17 Aug 1998 10:33:51 -0400 (EDT)

On Sat, 15 Aug 1998 Andries.Brouwer@cwi.nl wrote:
> After I sent a patch improving the speed of ext2, I hoped
> for people telling me about the effects it had. However,
> it seems all activity on the list is about killing X.

I did some benchmarks using the MySQL "sqlbench" benchmark using
2.0.34, 2.0.35 and 2.1.115. The machine is a dual Pentium II-400 with
512mb of memory, 12 9gb Seagate SCSI drives, and 2 4gb Seagate SCSI
drives. The benchmark was run on 6 of the 9gb Seagate SCSI drives
configured as a RAID-5 array using a ICP-Vortex GDT 5-channel RAID
controller. The GDT had a 64mb EDO SIMM cache onboard.

2.0.35 and 2.1.115 were compiled with SMP support. 2.0.34 was actually
2.0.34-pre6 or something like that, the stock Red Hat 5.1 kernel that
comes on the CD-ROM, and was not compiled with SMP support.

Here are the results:

Benchmark DBD suit: 1.2
Date of test: 1998-08-13 6:05:20
Running tests on: Linux 2.0.34 i686
Server version: MySQL 3.21.33 log
Totals per operation:
Operation seconds usr sys cpu tests
[...individual totals available upon request]
TOTALS 183 41.70 14.04 55.74 195725

Benchmark DBD suit: 1.2
Date of test: 1998-08-13 5:32:40
Running tests on: Linux 2.0.35 i686
Totals per operation:
Operation seconds usr sys cpu tests
[individual benchmarks available upon request]
TOTALS 222 20.17 58.06 78.23 195725

Benchmark DBD suit: 1.2
Date of test: 1998-08-13 6:19:32
Running tests on: Linux 2.1.115 i686
Totals per operation:
Operation seconds usr sys cpu tests
[...individual totals available upon request]
TOTALS 176 42.11 13.03 55.14 195725

Note that the machine was, of course, rebooted between tests (after all they
are under different kernels!).

It appears that SMP under 2.0 actually slows down I/O operations. 2.1
(SMP) was 4% faster than 2.0 (non-SMP) on this benchmark. It is
unclear whether improved dcache performance or improved filesystem performance
is to blame, but there is definitely a performance improvement.

The next step is to try 2.1 without SMP support. That's a bit more difficult,
due to the fact that Linus has a SMP machine. But one thing is clear -- 2.1
is definitely NOT slower than 2.0. At least where filesystem performance is
concerned.

Eric Lee Green eric@linux-hw.com Linux Hardware Solutions
Systems Specialist Quality Systems Designed For Linux

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html