About this list Date view Thread view Subject view Author view Attachment view

From: Micah Anderson (micah_at_riseup.net)
Date: Fri 08 Apr 2005 - 04:36:13 BST


Here is an update of more hacking done to port the vserver 1.9.5 patches
to the debian 2.6.8 kernel.

I've applied Bertl's second attempt at the patch ;) to the Debian
2.6.8-15 kernel-source tree, configured everything (including vserver
and kernel debugging options), compiled and installed the
kernel. Everything has been stable so far.

Tonight we tested a few of the TODO items:

. Uptime virtualization test

Uptime virtualization will cause the context to account all process
times and the uptime relative to the the context starting. When you do
not have the virtualized uptime flag set, the uptime in a context is
the host's uptime.

It was not expected that this test would work, but we wanted to know
for sure.

The test: chcontext --ctx 100 --flag ^17 bash -c "sleep 600 & sleep 500; uptime; ps auxwww"

This will change context to 100, set the uptime virtualization flag,
and then run two sleeps (the first in the background). If it is
working it should show something like 8 minutes uptime and the process
start time for the backgrounded sleep should be 8 minutes back, and
the current time for the "ps auxwww".

Result: The uptime worked, but the ps time was wrong

Bertl created new patches which should fix most of the issues
(possibly including the uptime), and incorporates all fixes waiting
for 1.9.6:

http://vserver.13thfloor.at/Experimental/patch-2.6.8-15-vs1.9.5.x-3.diff
http://vserver.13thfloor.at/Experimental/delta-2.6.8-15-vs1.9.5.x-2-3.diff

(second is a delta between the -2 and the first)

Three major things will not be included in this Debian version:

        - varhz: this changes the timer interrupt the kernel uses
          (was 100Hz for 2.4 and now is 1000Hz) the varhz allows you
          to change it between 20 and 20000 Hz, this reduces the
          slice time for interactive servers and/or decreases the
          scheduling overhead. It didn't make a really big difference
          so it probably will go away soon.

        - the memory split for x86: memory split simply doesn't work
          before 2.6.9*/10 (this allows to have other memory splits on
          i386 other than 3/1 which allows to have more than 1GB memory
          without himem. For example systems with 2GB memory using highmem
          for them will add significant overhead, avoiding that with a 2/2
          split is simply an advantage

        - bind mount extensions: http://lwn.net/Articles/124482/ bind
          bind mount extentions exist for 2.6.8, but are very different
          than the ones used for 2.6.11+, so they aren't included, if
          they are desired, then you should use the older patches, or
          move on to the 2.6.11 kernel

TODO:
        . Test the -3 version of the patches
         . Need to check task_state (in proc) regarding initpid
          virtualization (means /proc/<pid>/stat(us)
        . Need to check ext2/ext3 xattr with disk limit accounting
        . Need to test NFS
        . Need to test other basic capabilities
          (http://linux-vserver.org/Linux-VServer-Paper-14)
_______________________________________________
Vserver mailing list
Vserver_at_list.linux-vserver.org
http://list.linux-vserver.org/mailman/listinfo/vserver


About this list Date view Thread view Subject view Author view Attachment view
[Next/Previous Months] [Main vserver Project Homepage] [Howto Subscribe/Unsubscribe] [Paul Sladen's vserver stuff]
Generated on Fri 08 Apr 2005 - 04:36:44 BST by hypermail 2.1.3