Srija G
2010-Jul-03 21:57 UTC
[Xen-users] Problem pinging the xen guest after live migration
Hi , I am getting the delay in pinging (from different network) , the guest after live migration . Here is the scenario: 0. OS : RHEL5.4 86_64 1. Host-A. Port eth1 , it is trunked with three subnets xxx.xx.15.xxx xxx.xx.16.xxx xxx.xx.17.xxx Guest created on 15, 16 ,17 subnets. Gust15-A created on subnet 15 Gust16-A created on subnet 16 Gust17-A created on subnet 17 2. Host-B. Port eth0 and eth1 are bonded to bond0 Bond0 is trunked, with three subnets. xxx.xx.15.xxx xxx.xx.16.xxx xxx.xx.17.xxx Guest created on 15, 16 ,17 subnets. Gust15-B created on subnet 15 Gust16-B created on subnet 16 Gust17-B created on subnet 17 I have used the mode option in /etc/modprobe.conf as options bond0 mode=1 miimon=100 , As because mode=4 did not work, It seems the link aggregation is not properly set or something else. As the switch configuration is the part of different team, I can''t explain it clearly here. But with mode=1 , it is working at this stage. 3. I am trying to live migrate Guest15-A and other guests to Host-b. Migration is ok. But,there is 5 munites delay. The guests are pingable only after 4/5 munites from different network after migration. Can anyone please advice , where to check? Do you think it is the switch configuration issues? If server side configuration issue then I can do it. If it is switch side issues , please give me the pointers or advice , so that I can discuss with the network team. If more explanation is needed , please let me know. Thanks in advance. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Matej Zary
2010-Jul-05 10:05 UTC
Re: Re: [Xen-users] Problem pinging the xen guest after live migration
On Sat, 2010-07-03 at 17:57 -0400, Srija G wrote: Hi , I am getting the delay in pinging (from different network) , the guest after live migration . Here is the scenario: 0. OS : RHEL5.4 86_64 1. Host-A. Port eth1 , it is trunked with three subnets xxx.xx.15.xxx xxx.xx.16.xxx xxx.xx.17.xxx Guest created on 15, 16 ,17 subnets. Gust15-A created on subnet 15 Gust16-A created on subnet 16 Gust17-A created on subnet 17 2. Host-B. Port eth0 and eth1 are bonded to bond0 Bond0 is trunked, with three subnets. xxx.xx.15.xxx xxx.xx.16.xxx xxx.xx.17.xxx Guest created on 15, 16 ,17 subnets. Gust15-B created on subnet 15 Gust16-B created on subnet 16 Gust17-B created on subnet 17 I have used the mode option in /etc/modprobe.conf as options bond0 mode=1 miimon=100 , As because mode=4 did not work, It seems the link aggregation is not properly set or something else. As the switch configuration is the part of different team, I can't explain it clearly here. But with mode=1 , it is working at this stage. 3. I am trying to live migrate Guest15-A and other guests to Host-b. Migration is ok. But,there is 5 munites delay. The guests are pingable only after 4/5 munites from different network after migration. Can anyone please advice , where to check? Do you think it is the switch configuration issues? If server side configuration issue then I can do it. If it is switch side issues , please give me the pointers or advice , so that I can discuss with the network team. If more explanation is needed , please let me know. Thanks in advance. 1. Ping after 2. What happens if you issue the ping command from the _migrated_ domU and then tries pinging it from the another network? (the switch might "remember _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
John Haxby
2010-Jul-05 10:25 UTC
Re: [Xen-users] Problem pinging the xen guest after live migration
> On Sat, 2010-07-03 at 17:57 -0400, Srija G wrote: > Hi , > [snip] > Can anyone please advice , where to check? Do you think it is the switch configuration issues? > >It''s almost certainly a switch issue: the switch doesn''t know that the MAC address for the recently migrated machine is on a different port. There are two possibilities: you can either ping something (eg the default router) _from_ the guest after it has migrated or you can set portfast on the switch. The former just informs the switch that the originating MAC address is now on a different port; the latter makes the switch less willing to hold on to the MAC address/port association for a long time and is something that the network people can do. You might want to try doing both. jch _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Srija G
2010-Jul-07 15:27 UTC
Re: [Xen-users] Problem pinging the xen guest after live migration
Thank you so much to both of you John and Matej. Yes, both of your solution, worked when I am live migrating the guest from one dom0 to another dom0. What i am doing now, - In Host-B , I am executing the command , ping Guest-15a - In the Host-A , I am executing the command xm migrate --live Guest-15a Host-B The result: * Live migration went very well, found there is just 1 sec stop in between but I think that is acceptable.* Our network team says that portfast is already configured at switch . But as the first option worked , so I hope I can move on - right? If not please advice.... My second question is, mode=4 , ie 802.2ad ( link aggregation ) did not work still yet. I wanted to set the mode=4 as it is much faster. Found one document , here is the url http://wiki.oracle.com/page/Cisco+Systems+IOS-based+switches-+interface+bonding+and+trunking It is very good . In this document , the first line it is mentioned : ***** It is important that the native VLANs *be identical* on both sides of the link What does it mean VLANs *be identical* on both sides of the link ? Sorry for the silly question though. Would you please explain , do I need to check /configure anything on the server side? My configuration is simple enough , In my configuration I have used eth0 and eth1 bonded to bond0, and for each subnet, created the config files, as ifcfg-bond0.15, ifcfg-bond0.16, ifcfg-bond0.17. Then in the /etc/xen/scripts , I have created a custom file where I mentioned all the bridges for each vlan tag. And this custom file is being called from /etc/xen/xend-config.sxp... Please advice, do you need to proceed with aggregation link or with mode=1, which is now configured and you all helped me a work around... Thanks again. On Mon, Jul 5, 2010 at 6:25 AM, John Haxby <john.haxby@oracle.com> wrote:> > On Sat, 2010-07-03 at 17:57 -0400, Srija G wrote: >> Hi , >> [snip] >> >> Can anyone please advice , where to check? Do you think it is the switch >> configuration issues? >> >> >> > > It''s almost certainly a switch issue: the switch doesn''t know that the MAC > address for the recently migrated machine is on a different port. > > There are two possibilities: you can either ping something (eg the default > router) _from_ the guest after it has migrated or you can set portfast on > the switch. The former just informs the switch that the originating MAC > address is now on a different port; the latter makes the switch less willing > to hold on to the MAC address/port association for a long time and is > something that the network people can do. You might want to try doing both. > > jch > > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Bart Coninckx
2010-Jul-07 15:42 UTC
Re: [Xen-users] Problem pinging the xen guest after live migration
On Wednesday 07 July 2010 17:27:23 Srija G wrote:> Thank you so much to both of you John and Matej. > > Yes, both of your solution, worked when I am live migrating the guest > from one dom0 to another dom0. > > What i am doing now, > > - In Host-B , I am executing the command , > > ping Guest-15a > > - In the Host-A , I am executing the command > > xm migrate --live Guest-15a Host-B > > The result: > > * Live migration went very well, found there is just 1 sec > stop in between but I think that is acceptable.* > > Our network team says that portfast is already configured at switch . > > But as the first option worked , so I hope I can move on - right? If > not please advice.... > > My second question is, mode=4 , ie 802.2ad ( link aggregation ) did not > work still yet. I wanted to > set the mode=4 as it is much faster. Found one document , here is the url > > http://wiki.oracle.com/page/Cisco+Systems+IOS-based+switches-+interface+bon > ding+and+trunking > > It is very good . In this document , the first line it is mentioned : > > ***** It is important that the native VLANs *be identical* on both sides > of the link > > What does it mean VLANs *be identical* on both sides of the link ? Sorry > for the silly question though. > > Would you please explain , do I need to check /configure anything on the > server side? My configuration is simple enough , > > In my configuration I have used eth0 and eth1 bonded to bond0, and for > each subnet, created the > > config files, as ifcfg-bond0.15, ifcfg-bond0.16, ifcfg-bond0.17. Then > in the /etc/xen/scripts , I have created > > a custom file where I mentioned all the bridges for each vlan tag. And > this custom file is being called > > from /etc/xen/xend-config.sxp... > > > Please advice, do you need to proceed with aggregation link or with > mode=1, which is now configured and > you all helped me a work around... > > Thanks again. > > On Mon, Jul 5, 2010 at 6:25 AM, John Haxby <john.haxby@oracle.com> wrote: > > On Sat, 2010-07-03 at 17:57 -0400, Srija G wrote: > >> Hi , > >> [snip] > >> > >> Can anyone please advice , where to check? Do you think it is the > >> switch configuration issues? > > > > It''s almost certainly a switch issue: the switch doesn''t know that the > > MAC address for the recently migrated machine is on a different port. > > > > There are two possibilities: you can either ping something (eg the > > default router) _from_ the guest after it has migrated or you can set > > portfast on the switch. The former just informs the switch that the > > originating MAC address is now on a different port; the latter makes the > > switch less willing to hold on to the MAC address/port association for a > > long time and is something that the network people can do. You might > > want to try doing both. > > > > jch >Why don''t you use multipathing in stead of bonding? B. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users