Re: [sami] Welcome to SAMI and something you may like to know.

[Juergen Schoenwaelder <[e-mail]>]

On Thu, Aug 18, 2011 at 08:57:15PM +0800, Yingjie Gu(yingjie) wrote:

> The scope of State migration, i.e. the scope of VM migration, is
> within a Layer 2 subnet, which guarantees that VM's IP address can
> keep unchanged. Though some technologies can enable VM migrates
> between different subnets and keep the IP address unchanged, most
> requirements for VM migration is within the same L2 subnet for now.

Those setups that I have seen (but I have not seen too many and
usually smaller onces) have a single L2 network _behind_ firewalls and
load balancers and hence there is almost no state to migrate when a VM
moves. It appears to me that people design their networks to make
migration feasible (by making it a single L2 network) instead of
trying to migrate across arbitrary network topologies.

For me, the problem becomes interesting for the IETF when someone
wants to migrate a VM from one IP subnet to another IP subnet. In that
case, we have some form of IP mobility (since the IP connectivity
changes) and depending on the nature of the applications running on
the VMs, there might be some period in which some tunneling is needed
(because the state associated with existing transport connections in
middleboxes is likely harder to migrate correctly than doing some
mobile IP like tunneling). Some people said in Quebec that we can
ignore the IP addressing issues and just focus on the state migration
but I feel that is wrong because the way you deal with the mobility at
the IP level will have impact on how much state needs migration or
whether any operational state needs to be migrated at all (and my
personal preference would be to design the network to avoid the need
for state migration - that is to produce guidelines or best practices
to follow rather than producing more special purpose standards).

To what extend this is a real-world problem and to what extend vendors
of middle-boxes are interested to implement a standard in this area, I
do not know. But a good indication of interest is usually that some of
those vendors actively participate in the specification of a solution.

As stated in Quebec, to me this still sounds more like a research
effort than a standardization effort but I am happy to get convinced
otherwise.

/js

-- 
Juergen Schoenwaelder           Jacobs University Bremen gGmbH
Phone: [phone]         Campus Ring 1, 28759 Bremen, Germany
Fax:   [phone]         <http://www.jacobs-university.de/>