kernel performance issues 2.4.7 -> 2.4.17-pre8

From: Needham, Douglas (
Date: Thu Dec 13 2001 - 13:36:09 EST


        I wanted to create an automated performance test battery that I
could run against new kernels as a validation for the environments that I
work with.

What I did:

        I decided I would run dbench and tbench with increasing number of
instances from 1 to 10. I would also run Bonnie starting with the base
memory size, and then run it with twice the memory size and finally four
times the memory size of the machine. I tested various kernel
configurations. I have to have ipsec for the kernels I run, so all of these
kernels have the ipsec patches in them. However, since the Red Hat kernel
does not include the freeswan patches I brought down ipsec in all my test

Hardware used:

        A HP desktop with a Pentium III-500, 128M memory, 20 Gig harddrive.
The hardware was the same for all tests.

Software used:

        Dbench, tbench and bonnie.
        These tests were run immediately upon startup of the machine, so all
processes where the same for all tests.
        (This is an attempt to keep as many conditions the same as possible
while changing out only the kernel)


        In the results table the first column is the directory that I used
to store my results. The kernels that match these directories are :

Test-2.4.07-10-(timestamp) == Red Hat 7.2 base kernel
Test-2.4.09-13-(timestamp) == Red Hat 7.2 update kernel
Test-2.4.10ipsec-(timestamp) == 2.4.10 with freeswan patches
Test-2.4.14ipsec-(timestamp) == 2.4.14 with freeswan patches
Test-2.4.15ipsec-(timestamp) == 2.4.15 with freeswan patches (
and the umount pacth)
Test-2.4.16ipaup-(timestamp) == 2.4.16 with freeswan, aa, and
preemptable patches
Test-2.4.16ipsec-(timestamp) == 2.4.16 with freeswan patches
Test-2.4.16ipsp-(timestamp) == 2.4.16 with freeswan, and
preemptable patches
Test-2.4.17-pre8-(timestamp) == 2.4.16 with freeswan, and 17-pre8

What I found:

        Overall I discovered that the Red Hat modified kernel beat the stock
kernel hands down in throughput. Both the base Red Hat 7.2 kernel and the
7.2 update kernel (2.4.7-9, 2.4.9-13 respectively) had far better throughput
than the .10, .15, .14, .16, and .17-pre8 kernels.


        I have attached the results to this email as a gzipped html
attachment. It is simple a HTML TABLE that describes all of my results.

What I am looking for:

        Guidance. Could someone tell me what if anything I am doing wrong
with my tests, or point me in the direction to look where I can begin
understanding why the throughput performance dropped?




To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

This archive was generated by hypermail 2b29 : Sat Dec 15 2001 - 21:00:26 EST