Re: [Xen-users] Network dies when downloading on dom0

[Pepe Barbe <elventear@xxxxxxxxx>]

On Aug 24, 2008, at 8:39 PM, Truls Asheim wrote:

> The problem I'm having is that whenever i download a large file (3mb+)
> from the internet on the dom0 any networking on the host machine dies
> and the dom0 and all its domU's become completely unavailible for  
> about
> 5 minutes whereafter the network comes back up again and everything
> starts working.
>
> Does anyone have a clue of whats going on?
I have a similar setup on Ubuntu 8.0.4 and I am experiencing the same  
problem with the exact same symptoms.
What I have observed is that if the the rate of the incoming traffic  
is above certain threshold, ping times to the public internet starts  
to build up. So I am assuming there is some problem with VIF code,  
where too much traffic will incurr into backlogs until it fills up  
whatever queues there are. After the queues are full, you have to wait  
for a long while until everything is back to normal. The worse part,  
it that while the VIFs are backlogged and unresponsive, you cannot do  
something major like destroying the domUs or rebooting the dom0  
because some of the processes seem to be in uninterruptible sleep and  
then the entire system locks up warranting a cold reboot (Which in my  
case caused a resync of our RAID array, enough reason for me to avoid  
trying to troubleshoot this problem even more).
I have managed to mitigate the problem by limiting the transfer rate  
on all the VIF, using the 'rate' directive in all the config files.
Also, I have enabled QoS on Shorewall to limit the traffic rate on the  
public interface. In my case our connection is a 7Mbps/896kbps DSL  
line and by trial and error I determined that with a limit of 5Mbps I  
can avoid the DoS at least. I haven't been able to find a solution to  
this problem and get the full download rate of our DSL line.
Also, I don't know how well the rate limiting code for VIF in Xen  
works, because using that option alone, didn't fix much, just  
prolonged the amount of traffic that could go through the interface  
until it died.
Any ideas on how to solve this issue are welcome.

Pepe

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