2.6 ide i/o performance

2.6 ide i/o performance

[Vitalis Tiknius]

i simultaneously burned (audio mode, without scsi emulation) and ripped cd's 
under 2.6.0-test2-mm2. devices are:

 ../ide/host0/bus1/target0/lun0/cd (Teac CD-W552E) 
 ../ide/host0/bus1/target1/lun0/cd (Teac DV-516E).

mobo is Intel 875. software is k3b-0.9, cdrtools-2.01_alpha18, grip-3.1.1, and 
cdparanoia-3.9.8 with all paranoia options on.

when burning and ripping are performed separately, their speeds are approx. 
42x and 6.3x. when simultaneously, 12x and 1.6x with no options touched.

although devices are on the same controller (my first controller is SATA) and 
on the same bus, i'd rather expect linear and not almost square-law 
throughput regression observed. are the things expected to go this way, or 
there is some room for optimizations, etc.? 

Re: 2.6 ide i/o performance

[Måns Rullgård]

Vitalis Tiknius <vt@vt.fermentas.lt> writes:

> i simultaneously burned (audio mode, without scsi emulation) and ripped cd's 
> under 2.6.0-test2-mm2. devices are:
>
>  ../ide/host0/bus1/target0/lun0/cd (Teac CD-W552E) 
>  ../ide/host0/bus1/target1/lun0/cd (Teac DV-516E).
>
> mobo is Intel 875. software is k3b-0.9, cdrtools-2.01_alpha18, grip-3.1.1, and 
> cdparanoia-3.9.8 with all paranoia options on.
>
> when burning and ripping are performed separately, their speeds are approx. 
> 42x and 6.3x. when simultaneously, 12x and 1.6x with no options touched.
>
> although devices are on the same controller (my first controller is
> SATA) and on the same bus, i'd rather expect linear and not almost
> square-law throughput regression observed. are the things expected
> to go this way, or there is some room for optimizations, etc.?

The first thing to do is to use separate cables for those devices, and
see if it helps.

-- 
Måns Rullgård
mru@users.sf.net