On Sa, 10.07.2004, 17:36, Jon Burgess sagte: > Rene Bartsch wrote: > >>>I'm having artefacts in the picture (VDR). The HDD (SATA) performs with >>>58 MB/s, so I assume it's related to the PCI-bus (on an i865PE board). >>> >>> > My setup has an i865G running with 4 x Nova-T PCI cards and I don't see > any problems so I suspect the PCI bus is not the cause. >
Mine has three Novas (Satelco Standard DVB-S). > Does "femon" report any frontend errors when you see the artefacts? You > should be able to have femon running at the same time as VDR. > > Could you run "vmstat 1" while VDR is showing problems and reports a few > lines of the output? > I'll check that when I'm back home tomorrow. > I know of one problem when using a combination of VDR + Linux-2.6 + > ext2/3. This manifests itself in the vmstat output as a high CPU% as the > "io wait". > > Is this a combination that you use? > Yes, it is. Using a kernel 2.4.22 everything worked fine. Then I upgraded to 2.4.(don't know wether it was 24 or 26). This caused problems with the buggy S-ATA/P-ATA modules. So I decided to directly go to 2.6.6 using libata. FS is Ext3. But the artefacts also appear when streaming (streamdev-server). > VDR triggers a preformance regression in the 2.6 kernel by streaming > lots of small writes to the filesystem. > I did some benchmarks to demonstrate the problem a few months ago, but > AFAIK it has not been fixed. > The three week old iptables-bug hasn`t been fixed, too (even Debian doesn't provide patched kernels, yet). Things currently seem to slow down in kernel development. > To "fix" the problem I changed my video storage drive to use JFS. The > benchmarking I performed showed that JFS, XFS and ReiserFS v4 are not > effected by this issue. > Ok, I'll check vmstat and CPU load (although it shouldn't be a problem on a HT-machine?). Would be glad if changing FS helps :-) Thanx :-) Rene
