george.shuklin at gmail
Mar 20, 2012, 6:43 PM
Post #2 of 2
On 20.03.2012 23:35, Brendan Morrison wrote:
> Hello all, I have recently finished a project researching solutions to
> perform virtual machine migration across ip subnets
> while maintaining tcp/ip connections. I am currently looking into
> continuing this work at a masters level and/ or using it for the
> google summer of code under either organization debian, or Gentoo.
> Anyways the current virtualization tecknology used is KVM and it is a
> limiting factor in our design. As such I am trying to look into other
> free open source technologies I can use for virtualization.
> Basically I am wondering the current capabilities of XEN mainly for
> live migration purposes and more specifically if there is any
> capability that would allow me to migrate a VM while it is running and
> upon completion of migration be left with 2 instances of the VM running?
> I understand that the two machines will obviously not be synchronized
> .. unless thats possible too? But I am woundering what current built
> in capabilities are as I would much rather focus on the networking
> side of the problem and limit the hacking to the
> current migration technologies to improve the design as this
> networking is more to where my expertise reside.
> Thanks in advance
> xen-api mailing list
> xen-api [at] lists
I see no specific problem with this. Any migration at the end is just
'create new domain', 'copy', 'destroy old domain'. You can modify xapi
sources or xen sources not to wait death of original domain, but just
change it uuid to 'technical' to avoid confusion. The main problem is
that migration is usually solution to free node for maintenance, so
'leftover' domain will die...
I thought about way to do that type of migration, and the simplest way
I found (without modifying toolstack or deep changing in guest domain)
is just use a /32 networks (IPv4). In that case all packets will be
routed via gateway (outside designated network). During migration route
will be just replaced with new route table in guest domain and router(s).
If you do that, all traffic (except few dropped packets, which is not
problem for TCP) will be routed to new designation and accepted by guest.