This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
Home Products Support Community News


Re: [forums] Re: [Xen-users] domU network fails under load - vif breaks

To: xen-users@xxxxxxxxxxxxxxxxxxx
Subject: Re: [forums] Re: [Xen-users] domU network fails under load - vif breaks
From: Paul <xensource@xxxxxxxxxxx>
Date: Fri, 12 Sep 2008 08:27:39 +1000
Delivery-date: Thu, 11 Sep 2008 15:28:20 -0700
Envelope-to: www-data@xxxxxxxxxxxxxxxxxxx
In-reply-to: <70c5648c0809111134t4198529ap687a5905943fc323@xxxxxxxxxxxxxx>
List-help: <mailto:xen-users-request@lists.xensource.com?subject=help>
List-id: Xen user discussion <xen-users.lists.xensource.com>
List-post: <mailto:xen-users@lists.xensource.com>
List-subscribe: <http://lists.xensource.com/mailman/listinfo/xen-users>, <mailto:xen-users-request@lists.xensource.com?subject=subscribe>
List-unsubscribe: <http://lists.xensource.com/mailman/listinfo/xen-users>, <mailto:xen-users-request@lists.xensource.com?subject=unsubscribe>
References: <70c5648c0809111134t4198529ap687a5905943fc323@xxxxxxxxxxxxxx>
Sender: xen-users-bounces@xxxxxxxxxxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-AU; rv: Gecko/20080829 Lightning/0.8 Thunderbird/ Mnenhy/
I have been running for a few days now with a tulip 10/100 card, and without issues. The original realtek was a gigabit card, so this isn't a one for one test - however, I was getting the network breaking at 3-4Mbs as far as I could tell. I am using nfs to move the data, and this is directly from domU to dom0 - I haven't fully got my head around whether this traffic even traverses the nic - does xen have the smarts to keep the traffic on the vif? Any, the same action was causing a break, and is no longer.

Is the vif speed the same as the physical nic? ethtool only shows link up from the domU's perspective of the card.

I am going to source a non-realtek gigabit card and see if the problem is just a combination of gigabit and gentoo somehow.

Jon Simonds wrote:
Nathan Eisenberg wrote:
I'm wondering if #3 is a large enough change to rule out driver issues.  Maybe 
try an entirely different chipset and manufacturer - perhaps an Intel, 
Broadcom, etc NIC.

I'm curious if anybody has made any further progress on this.
I too have the exact same issues as Paul-372 (identical.) Only fix is
to reboot the dom0.
Now running Xen 3.3.0 with Xen-Sources 2.6.21 on a gentoo base system.
Same issues on Xen 3.2.0.

I will add to the list that I am using a SuperMicro based server, with
built-in Intel Pro/1000 Nics.
I've had this issue from the beginning, and have tried to limit things
that cause the network bridge to collapse, but this makes the machine
worthless.  This is supposed to be a production level machine for us,
but at the moment it obviously isn't useable.  I can cause an
immediate drop of all the vifs by doing a copy to an nfs share on dom0
from any of the domU machines.  I am running 4 PV domUs and 1 Win32
DomU on this machine.  I will provide any information that may help in
determining the root of the problem.

from lspci:
0d:00.0 Ethernet controller: Intel Corporation 82573E Gigabit Ethernet
Controller (Copper) (rev 03)
0e:00.0 Ethernet controller: Intel Corporation 82573L Gigabit Ethernet

from syslog:
Sep 10 18:58:15 xen Intel(R) PRO/1000 Network Driver - version 7.3.20-k2
Sep 10 18:58:15 xen e1000: eth0: e1000_probe: Intel(R) PRO/1000
Network Connection
Sep 10 18:58:15 xen e1000: eth1: e1000_probe: Intel(R) PRO/1000
Network Connection

I see that intel has a new driver, but it is only available as a
module.  I tend to compile the kernel for exactly what I need, and
keep the machines as lean as possible.  I would prefer not to go this
route, but will if it offers a fix.  I have not done this yet, as it
seems to not matter what NIC is being used.  I believe I may be the
first to post about the problem with Intel Pro/1000 nics.

So far for me, this problem seems to be strictly related to heavy
traffic between the dom0 and the domUs.  If I place an nfs server on
one of the other domUs I seem to be able to transfer files without
causing the vif drop.  I have a colleague that has the exact same
servers, and he doesn't do any domU to dom0 traffic, just domU to the
internet, and he is able to sustain from 15Mb/sec to 100Mb/sec data
streams without dropping the vifs.  Maybe it is possible that this
level of throughput is not causing him problems, as I'm not completely
aware of what type of throughput between my DomUs and Dom0 is actually
dropping the vifs, but I see that it is as low as 3Mbit for some

-- I appologize if this hits the list incorrectly or misses the
thread... I tried to reply through Nabble, but my post has been
pending since yesterday, so I figured I would try posting directly to
the list.

Xen-users mailing list

*email: paul@xxxxxxxxxxx | msn: paul@xxxxxxxxxxx | gtalk: paul@xxxxxxxxxxxxx | skype: paulkoan | y!: paulxkoan | blog: http://www.servwise.com/blog-paul*
*ServWise Advanced Hosting www.servwise.com <http://www.servwise.com> *Standard and Reseller web hosting on Windows and Linux* *

Xen-users mailing list