Klaus Schmidinger wrote: > Emil Naepflein wrote: > >On Mon, 10 Oct 2005 09:54:56 +0300, Marko M?kel? wrote: > > > >>On Mon, Oct 10, 2005 at 06:52:12AM +0200, Emil Naepflein wrote: > >> > >>>This is pretty useless on my system because the loss of responsiveness > >>>is mostly caused by the disk accesses and not by cpu usage. > >> > >>Are you sure that DMA disk transfers are enabled? If not, the copying > >>with programmed I/O (PIO) will consume CPU. > > > > > >As I wrote, there is no problem with cpu usage. And DMA is enabled as I > >use a Promise SX6000 controller (SCSI interface). The main problem is > >that the cutting thread reads and writes the data through the buffer > >cache and purges useful data from the buffer cache. Any disk access when > >using the menu then is delayed considerably. This is with a 2.4 kernel. > >The 2.6 kernel behaves much better in this regard. > > I'm currently in the process of adopting Ralf M?ller's patch > to avoid trashing the disk cache. Maybe that will help. 2.6.13+ also supports ionice. That could help to not starve other processes accessing the disk. AFAIK it's on a per process level though so the cutting thread would need to be turned into to a separate cutting process. cu Ludwig -- (o_ Ludwig.Nussel@xxxxxx //\ V_/_ PGP Key ID: FF8135CE
