Patch or module?

Patch or module?

[cschumpf]

If thats not the right group to discuss that, please let me know.

I would like to write an IO-Bandwidth-Limiter on per Process and per
User-Basis for a few disk drives. I can either patch the kernel functions read/write
and enhance the task- and user-structure and globally check if the correct
devices are adressed or I can write my own module, twist pointers from the
filesystems on the drives and store the information about users and tasks there.

I like the second approach way better, but its maybe slower. I have to do an
extra search in my structures for user and task ids, which is maybe more
expensive than an 'if' in the read/write-funktion.

Any guidelines? Any recommendations?

Thanks


-- 
GMX - Die Kommunikationsplattform im Internet.
http://www.gmx.net

Re: Patch or module?

[Oliver Neukum]

On Tuesday 19 February 2002 19:23, cschumpf@gmx.net wrote:
> If thats not the right group to discuss that, please let me know.
>
> I would like to write an IO-Bandwidth-Limiter on per Process and per
> User-Basis for a few disk drives. I can either patch the kernel functions
> read/write and enhance the task- and user-structure and globally check if
> the correct devices are adressed or I can write my own module, twist
> pointers from the filesystems on the drives and store the information about
> users and tasks there.

You cannot do this on the device level as the writing task is not identical
to the task actual having caused the requirement to do IO.

	HTH
		Oliver

Patch or module?

[cschumpf]

If thats not the right group to discuss that, please let me know.

I would like to write an IO-Bandwidth-Limiter on per Process and per
User-Basis for a few disk drives. I can either patch the kernel functions read/write
and enhance the task- and user-structure and globally check if the correct
devices are adressed or I can write my own module, twist pointers from the
filesystems on the drives and store the information about users and tasks there.

I like the second approach way better, but its maybe slower. I have to do an
extra search in my structures for user and task ids, which is maybe more
expensive than an 'if' in the read/write-funktion.

Any guidelines? Any recommendations?

Thanks


-- 
GMX - Die Kommunikationsplattform im Internet.
http://www.gmx.net

Patch or module?

[cschumpf]

If thats not the right group to discuss that, please let me know.

I would like to write an IO-Bandwidth-Limiter on per Process and per
User-Basis for a few disk drives. I can either patch the kernel functions read/write
and enhance the task- and user-structure and globally check if the correct
devices are adressed or I can write my own module, twist pointers from the
filesystems on the drives and store the information about users and tasks there.

I like the second approach way better, but its maybe slower. I have to do an
extra search in my structures for user and task ids, which is maybe more
expensive than an 'if' in the read/write-funktion.

Any guidelines? Any recommendations?

Thanks


-- 
GMX - Die Kommunikationsplattform im Internet.
http://www.gmx.net