N2100 Linux kernel write coalescing bug

[adamp]

Did anyone manage to create a module which enables write coalescing (that's apparently disabled by the boot loader)?

More info:
N2100 Linux kernel write coalescing bug
( From Thecus Wiki )

A tiny bug has been uncovered in the linux kernel that, when fixed, promises to increase system by up to a factor of two.

It is certainly present in linux versions 2.6.9 to 2.6.24. See Peter Teichmann's initial report and Peter Horton's confermation later in the thread.

lists.arm.linux.org.uk/lurker/message/20080302.122143.6b50e334.en.html

Kernel patch:
lists.arm.linux.org.uk/lurker/message/20080309.171257.ea11eb8d.en.html#linux-arm-kernel

[wengi]

Hi,

i am pushing this because it is a very interresting post!
If i am getting this right: This is a general bug in the kernel.

Thecus original firmware AND debian installations are affected, right?

Background: I think about installing debian on all N2100 but there was a poor io speed reported for debian several times.
Maybe this will be solved by the kernel patch.

wengi

[adamp]

Mar 25, 2008 15:38:14 GMT 7 wengi said:

Hi,

i am pushing this because it is a very interresting post!
If i am getting this right: This is a general bug in the kernel.

Thecus original firmware AND debian installations are affected, right?

Background: I think about installing debian on all N2100 but there was a poor io speed reported for debian several times.
Maybe this will be solved by the kernel patch.

wengi

The guy thought it was a kernel bug. But the problem is actually caused by the bootloader (Redboot) turning off write coalescing bit for some reason.

[markus]

Very interesting thread! What exactly can I do now to speed up my N2100 system? Is there anything to be executed with ssh (e.g. running a script)? And, if yes, how to do it?
kind regards
Markus

[majority]

Any updates on this?

I went through the mailing list and it looks like in this mail that Peter Horton made a module for his n2100 that fixed the issue.

Have I misread, or someone want to write to him and hear if he'd like to share - if the module is easy enough for the general public to be able to use? ;D

majority.

[getmythe]

I compiled a new kernel with the proposed patch

The performance gain is rather disappointing. It's somewhere between 0 and 2 percent only (e.g. within the margin of error for the measurements I did). Cannot distribute the kernel though since it is missing hooks for two wireless drivers. The Thecus kernel sources seem to be incomplete.

getmythe

[majority]

Ooo, all right. That does sound dissapointing. Oh well, life goes on, I shall continue to be happy with my n2100 as it is .

majority