Re: does anyone run guests for more than 5 minutes? (virtio-net perf anomaly)

[Marcelo Tosatti <mtosatti@redhat.com>]

On Tue, Mar 03, 2009 at 01:13:41PM -0700, Alex Williamson wrote:
> It seems like something happens around the 5 minute uptime in the guest
> that causes virtio-net throughput to plummet.  Here's the scenario:
> 
> guest started as:
> 
>         taskset -c 4 /usr/local/bin/qemu-system-x86_64 -hda /dev/sdb -m
>         2048 -vnc :1 -net nic,macaddr=02:00:10:91:73:02,model=virtio
>         -net tap,script=$HOME/bin/null-ifup -serial
>         tcp::1234,server,nowait -mem-path /hugepages/
> 
> null-ifup looks like this:
> 
>         #!/bin/sh
>         /sbin/ifconfig $1 192.168.0.1
>         /sbin/route add -net 192.168.0.0/24 gw 192.168.0.1
> 
> The guest gets a static IP of 192.168.0.2.
> 
> netserver (part of netperf) in the host is pinned to CPU0, which shares
> cache with CPU4 from the above taskset.
> 
> When the guest boots, I run:
> 
>         netperf -c -C -H 192.168.0.1 -t TCP_STREAM -- -m 64k
> 
> This results in ~13.5Gbps (note you won't get close to this if you don't
> get the tasksets correct)
> 
> Wait 5 minutes, retry.  Now I get ~4Gbps.  The only way I can get
> 13.5Gbps again is by rebooting the guest within the same qemu context,
> or of course restarting it completely.
> 
> Any guesses as to what might be going on?  Can anyone reproduce?  I'm
> hoping that I'm doing something dumb, but can't figure out what it is.
> The system is running v2.6.29-rc6-121-g64e7130 in the guest,
> v2.6.29-rc6-123-gbd7b3b4 on the host, kvm module kvm-84-620-g5bffffc and
> userspace kvm-84-95-gea1b668.  Thanks,

Nope. Collect kvm_stat -l before/after the slowdown?