Network poblems using virtio and tap.

["Fabian Deutsch" <fabian.deutsch@gmx.de>]

Hey.

I'm observing some network problems when using virtio and a tap if.
The trouble is a drop of network troughput from abot 90Mbit/s to about 90kbit/s,
after about 1.7 TB of transferred data.

Setup
Host: 
- Fedora 8, nearly all updates
- Intel(R) Xeon(R) CPU           X3210  @ 2.13GHz
- 8GB RAM
- 2 Guests

Guest:
- Fedora 9 all updates
- virtio nic using tap
- samba sharing a mountpoint using an iscsi target

Steps to reproduce:
1. Setup a guets using virtio and tap
2. Mount an iscsi target and wokr on the fs or  create some traffic on the nic.
2.a I used:
# while true ; do sync ; \
    time dd if=/dev/zero of=test.bin bs=1M count=1024 ; sync ; 
    time md5sum test.bin ; done
3. Wait a couple of days or about 1.7 TB of data on a 100Mbit/s.

Expected results:
Everything runs fine, even after 2.5 TB of transferred data or more.

Actual results:
Everything runs fine up to some point. 
Then: Troughput drops a bit (from 11 MB/s to about 6MB/s). 
At some point later the troughput drops to about 90kBit/s.
No messages on guest and host. No errors on ifconfig, no dmesg errors.
Ping latency increases over time. Logged here: 
http://www.meinsandkasten.de/log-virtio.pdf


Did someone observ something similar or so?

Greetings
fabian
-- 
Fabian Deutsch <fabian.deutsch@gmx.de>