[Xen-users] Jumbo frames with domUs in bridging mode

[Trevor Bentley <trevor.bentley@xxxxxxxxxxxxxxxxxxx>]

Hi everyone,

I have looked through the list archives and seen a few posts about jumbo 
frame support, but there is no definitive answer on whether or not it 
has been tested or is working for anyone.  I tried using an MTU of 9000 
in a domU and encountered some very interesting and not-so-great 
results.  My machines used in this test are:
* NFS (physical)
* dom0 (physical)
* domU (virtual)

The domU boots using the kernel's built-in NFS-root, with its filesystem 
stored on the NFS machine.  I set the MTU to 9000 on eth1, peth1, and 
vif0.1  (on the physical machines) on the three machines, and the xenbr1 
bridge on dom0 correctly displays MTU 9000 now.  I used "ping -s 8000 
<ip>" and "tracepath <ip>" to determine connectivity between the 
machines.  The two physical servers, NFS and dom0, work completely and 
can send 9000 byte packets back and forth.  However, domU was not so 
lucky.  I can send 9000 byte packets *out* of domU without a problem to 
both the NFS server and the dom0.  However, coming back in the maximum 
ping packet size it accepts is 4035 (i.e. "ping -s 4035 <ip>" works, 
"ping -s 4036 <ip>" hangs forever).  I don't know how big the headers 
are, but I wouldn't be surprised if the actual limit is 4096.
I'm using CentOS's default package for the hypervisor and dom0 kernel, 
and Debian Lenny's package for the domU kernel.  That means the 
hypervisor is 3.0.3 with most of the 3.1 changes backported, I believe.  
I'm not sure where this problem would be, though... since it's traveling 
through the bridge correctly in one direction, I think it's safe to 
assume the bridge is fine.  That makes me think it's either Xen's 
netfront or netback driver.
My questions to all of you are:
 1) Has anyone else encountered this, and do you know how to fix it?
 2) Is anyone using jumbo frames in their domUs, and is it reliable?

Thank you much,

Trevor

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