Virajith Jalaparti
2011-Jan-12 02:51 UTC
[Xen-devel] Live migration-delay in gratuitous arp
Hi, I was trying to perform live migration with a VM on a LAN and measure the downtime seen by an application (e.g. ping), the time for which the application cannot communicate with the VM. The application is run on a machine different from those between which the VM is migrated. I noticed that the VM remains unreachable for nearly 1-1.5sec. However the xend.log file says that the time (the delta) for the last round of the migration is around 100msec. So, the VM is actually paused for just around 100msec but the application cannot reach it for 1-1.5sec. I believe this is because of the delay in the gratuitous ARP being sent. Is the implementation such that the ARP is not sent immediately after the migration is finished? Is there an easy tweak (without modifying source code) by which I can correct this or do I have to modify the source code? I am using Xen from the repos of CentOS 5.5 on a 64bit machine. Thanks, Virajith _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jamon Camisso
2011-Jan-12 08:04 UTC
Re: [Xen-users] Live migration-delay in gratuitous arp
On 1/11/2011 9:51 PM, Virajith Jalaparti wrote:> Hi, > > I was trying to perform live migration with a VM on a LAN and measure > the downtime seen by an application (e.g. ping), the time for which the > application cannot communicate with the VM. The application is run on a > machine different from those between which the VM is migrated. I noticed > that the VM remains unreachable for nearly 1-1.5sec. However the > xend.log file says that the time (the delta) for the last round of the > migration is around 100msec. So, the VM is actually paused for just > around 100msec but the application cannot reach it for 1-1.5sec. I > believe this is because of the delay in the gratuitous ARP being sent. > Is the implementation such that the ARP is not sent immediately after > the migration is finished? Is there an easy tweak (without modifying > source code) by which I can correct this or do I have to modify the > source code? > > I am using Xen from the repos of CentOS 5.5 on a 64bit machine.I''d be interested to hear if running arping in screen during migration on your domU has any effect. Jamon _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Virajith Jalaparti
2011-Jan-12 16:46 UTC
[Xen-users] Re: Live migration-delay in gratuitous arp
I''d be interested to hear if running arping in screen during migration on your domU has any effect. Jamon Hi Jamon, I did not see any difference even if I do this. I ran "arping -I eth0 -b -U<domU ip address>" but still was getting a downtime of around 1.5sec. I also tried arping from one of the dom0''s but that does not help either. Virajith _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Jordan Pittier
2011-Jan-13 17:53 UTC
[Xen-users] Re: Live migration-delay in gratuitous arp
Hello, Try to run the following script from *another* dom0 in the same LAN. This script could be considered as an "hack" to update the CAM table of the switch : #!/bin/bash if [ -z "$1" ]; then echo "Error: give the DomU IP to ping as parameter." exit 1 fi HOSTNAME=$1 while /bin/true;do /sbin/ip neigh flush all &>/dev/null ping -c1 -w1 $HOSTNAME sleep 0.2 done Jordan _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Virajith Jalaparti
2011-Jan-13 19:26 UTC
Re: [Xen-users] Re: Live migration-delay in gratuitous arp
I had tried something similar earlier but it did not work. When even I run the script you sent, the atleast 10-15 of the consecutive pings don''t receive any replies. I guess this implies that some thing is wrong with the ARP sending mechanism of the VM. Apparently, this issue was discussed earlier at http://lists.xensource.com/archives/html/xen-devel/2007-03/msg00268.html and they say that this is a bug in Xen-3.0.3 which is the version in the repos of CentOS 5.5. Virajith Jordan Pittier wrote:> Hello, > Try to run the following script from *another* dom0 in the same LAN. > This script could be considered as an "hack" to update the CAM table > of the switch : > > #!/bin/bash > if [ -z "$1" ]; then > echo "Error: give the DomU IP to ping as parameter." > exit 1 > fi > HOSTNAME=$1 > while /bin/true;do > /sbin/ip neigh flush all &>/dev/null > ping -c1 -w1 $HOSTNAME > sleep 0.2 > done > > Jordan > > > ------------------------------------------------------------------------ > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users