Xen users - May 2008 - About the time of live migration, it is too long

Hi,all
I am currently doing some research about the live migration feature of xen.
It is said that the downtime can be less than 1 sec. But I can not get this
result. I being the live migration ,and "ping" the guest IP,and found
the
total down is more than 30 seconds. Than I tried migration without live it
only stop about 15 minutes!
So I am really puzzled about the result, why this happen?

By the way, my system configuration is:
The source server: cpu:Intel(R) Xeon(TM) CPU 2.40GHz Mem:512M(I also try
with Intel(R) Pentium(R) D CPU 2.80GHz Mem 512M, the same as the dest
server)
The dest server: Intel(R) Pentium(R) D CPU 2.80GHz Mem 512M
the guest os:Mem 128
all the operation system kernel is 2.6.18, the xen is 3.1.1, installed with
centos5
All the machine is interconnected by a 1 Gbps Ethernet switch

I am not sure am I describe my situation clearly. Any question about my
environment is welcome.
And any suggestion is welcome, thanks.

-- 
Yin Xu(ronee)
Information Security
Department of Computing and Information Technology
Fudan University
Shanghai, P.R.C.


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On Thu, 29 May 2008, Yin Xu wrote:
> So I am really puzzled about the result, why this happen?
Did you check for ARP caches?


Stefan




_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Yin Xu schrieb:
> Hi,all
> I am currently doing some research about the live migration feature of 
> xen. It is said that the downtime can be less than 1 sec. But I can not 
> get this result. I being the live migration ,and "ping" the guest
IP,and
> found the total down is more than 30 seconds. > Than I tried migration 
> without live it only stop about 15 minutes!
> So I am really puzzled about the result, why this happen?
ARP tables are perhaps not updated on your switches; that''s why you 
don''t get any replies to ping.

Try either (or both):
- using arping instead of ping,
- start "ping" command from within your guest (i.e., some other host
in
the network), then do a live migration.

Did the migration time decrease?


As of "without live" migration - as I understand you correctly - did
it
take fifteen minutes for you? Either your guest has to be very big (lots 
of RAM), your network is very slow (packet losses etc. - check the speed 
of migration with a tool like iptraf or bwm).


-- 
Tomasz Chmielewski
http://wpkg.org


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Hi Yin,

What kind of application is running when you do your live-migration? Very
CPU-intensive stuff?
Xen makes use of an iterative pre-copy mechanism by looking at the "dirtied
pages" in the RAM and retransferring these pages. Then it suspends the VM
on the one machine, transfers the "hot pages" that change very
frequently and then restarts the VM on the destination machine. But since your
RAM is not that big and since you use a GigE connection, there should be no
problem. I would assume this is related to the switch somehow. If you want to
eliminate this and just watch it for research, simply use a cross-over cable (if
the machines are close enough to each other ;)). In my test I had no
ping-loss...
You could also have a look at both machines with the command "watch -n 0.1
xm list" to see when the machine is ready.

Hope that helps!

Best regards,
Bjoern
_________________________________________________________________________
In 5 Schritten zur eigenen Homepage. Jetzt Domain sichern und gestalten! 
Nur 3,99 EUR/Monat! http://www.maildomain.web.de/?mc=021114


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users