[Xen-devel] Wonky results with pvops live migration under 4.0.1

[Pim van Riezen <pi+lists@xxxxxxxxxxxx>]

Good day,

We migrated one of our Xen clusters to the gitco 4.0.1 release after seeing 
that it seemed to fix the live migration issues we were having under 3.4.3 on 
our test cluster. It turns out our tests had no bearing on our production 
cluster.

The production cluster is a mix of mostly AMD Opteron with a few Intel Xeon. 
Migration issues do not correlate to cpu architectures, however.

I have run a test with a machine running pvops 2.6.32.25 performing live 
migrations. For some source & destination pairs the migration would 
consistently work. For other combinations it would consistently break.

The symptoms when it breaks are that the console does not respond to keyboard 
input. The network does ping. Interactive ssh sessions no longer work. The 
shell is loaded, but it also does not respond to keyboard input. Noninteractive 
ssh calls are normally executed. If a shutdown is sent to the vps, the shutdown 
sequence seems to hang on running sync.

All dom0 nodes run CentOS 5.5 with gitco Xen 4.0.1 repositories. Here is a raw 
dump of some migration tests:

A = 16 cores Xeon, 48GB memory, Emulex FC 8Gb
B-E = 24 cores Opteron, 128GB memory, Emulex FC 8Gb

C->B    OK
B->C    FAIL
B->C    FAIL
C->B    OK
D->C    OK
C->D    FAIL
C->D    FAIL
B->D    FAIL
D->B    OK
E->F    OK
F->E    FAIL
F->E    FAIL
E->A    OK
A->E    FAIL

The guest has 256MB memory, all dom0s are configured in bridging mode (with the 
bridges attached to a vlan interface). 

Is there anything I should try?

Cheers,
Pim van Riezen


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel