On 16.04.2012 00:41, Hans-Kristian Bakke wrote: >> As for the network-related issues, indeed, this is something which >> is quite unexpected. I browsed all changes between 1.0 and 1.0.1, >> but I don't see a single change which may have this effect... > > I am no dev but I would guess that several gbit/s of TCP-traffic may > trigger any regressions in other parts of the KVM subsystems too, so > perhaps it isn't directly linked to the networking?
I built 4 executables on current wheezy/amd64, available at http://www.corpit.ru/mjt/tmp/kvm1.0-network-speed-tests.tar.gz (and signed with my Debian gpg key, in http://www.corpit.ru/mjt/tmp/kvm1.0-network-speed-tests.tar.gz.asc -- the same key I use to sign my regular uploads to Debian archive, it should be in your apt keyring). These files are, obviously, temorary.. ;) The tarball contains the following files: kvm1.0 kvm1.0+2061800b85ddcc9b34b5ccbfaa87f7e8b94626a6 kvm1.0+a25808dc5baee83f36e0cdab998eb6c0024156fa+7e2191ae9898cc957a3d1991aff0e40f2e0f44a4 kvm1.0+fe5c13ebf1161d0f324229cfb36cb5fb87ec6248 These are /usr/bin/kvm binaries. So you extract these to some temp place and run your guest(s) using one of the 4 binaries instead of /usr/bin/kvm. kvm1.0 should be fast, the same as 1.0+dfsg-8 (I need a verification of this, too, as a starting point). The other 3 are 3 possible -- very remotely but still -- changes from between 1.0-8 and 1.0-9 which may show the bad speed. The long hex numbers are GIT commit IDs, which can be looked at git.qemu.org, like this http://git.qemu.org/?p=qemu.git;a=commit;h=COMMITID f.e. for the first one above: http://git.qemu.org/?p=qemu.git;a=commit;h=2061800b85ddcc9b34b5ccbfaa87f7e8b94626a6 (the pre-last is actually 2 but related commits). Please test them and see if the speed drops with any of them. As I already pointed out, all changes appears to be unrelated to networking, so these are wery wild guesses only, and neither may show our bad behavour. And especially if first, simple, kvm1.0 variant _does_ show the bad speed already, there's no need to try others, since the prob might be somewhere else entirely: this binary is equivalent to what was in 1.0-8 but rebuilt using current toolchain from wheezy. Thank you very much! /mjt -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
