aarrgh. sorry I ment to post this: Hi,
I was asked off-list to gather some more data, which I now present to anyone who's interested. Disclaimer: there is no acute problem to fix, but something is odd. Summary: - the location of a tgz which includes an acpidump and some dmesgs is: <http://stuff.ghweb.de/h8ssli2/stuff.tgz> - I tested the following kernels on a Supermicro H8SSL-i2 with an Athlon64 X2: amd64: GENERIC, GENERIC.MP, i386: GENERIC, GENERIC.MP - all except the i386/MP kernel have about 70% cpu load on interrupts (the mp kernels on one core), regardless whether acpi is enabled or not on an otherwise idle standard installation (up to 2 ssh-sessions active), and up to 90% interrupt load, if they've got something to do. - the i386/MP kernel has about 0.0% interrupt load, also regardless whether acpi is enabled - I did the following to put some load on I/O: a) ping -f from another machine to this (only 100 MBit-Network), with no packet loss in every case b) dd if=/dev/zero of=/tmp/stuff bs=1m count=2000 - the interrupt count monitored through systat vm 1 is roughly comparatively on all kernels, with and without acpi, except for pciide, which comes short on UP kernels (both count and transfer rate) Some data (interrupt count with systat vm 1): idle: amd64/MP amd64/UP i386/MP i386/UP clock 200 100 200 100 ipi 100 rtc 128 128 bge0 5+/-3 5+/-3 5+/-3 5+/-3 a) (ping-f) (clock, ipi and rtc same as above) bge0 3.4k 3.3k 3.3k 3.4k +/-100 +/-50 +/-50 +/-50 a) + b) (ping and dd) (clock, ipi and rtc same as above) bge0 3.4k 3.3k 3.3k 3.4k +/-100 +/-50 +/-50 +/-50 peaks of peaks of up to +2k up to +2k pciide 3.7k 850 3.4k 530 +/-150 +/-50 +/-500 +/-20 overall data transfer with dd: amd64/MP: 53-58 MB/s amd64/UP: 11-12 MB/s (about 19MB/s without ping -f) i386/MP: 52-56 MB/s i386/UP: 8- 9 MB/s --knitti