Re: [sami] 答复: First SAMI email in the new year, can we go further? Look forward to your opinions.

[Juergen Schoenwaelder <[e-mail]>]

On Fri, Mar 02, 2012 at 02:31:53AM +0000, Dacheng Zhang(Dacheng) wrote:
> >> 
> >> Lets assume VM migration happens today (which I am fine to assume - I
> >> am however unsure how frequent live migration takes place and hence
> >> whether its worth to optimize for it).
> [Dacheng Zhang] I think this will happen quite frequently in some conditions. Assume for energy conversation purpose at night an operator of a DC can move active VMs to several physical servers and switch other physical servers off. In this case, migration will happen very day.

Sure - but this still does not convince me there is a problem worth
solving. As long as there is a common layer two infrastructure,
migration seems to work today. If there is not a common layer two
infrastructure, you still need to have a common shared storage system.
I know, you will now draw a picture that VM will migrate across DCs
etc. We have discussed all these potential use cases. I am lacking
evidence that they (a) are real and (b) significant enough to be worth
solving. To me, this still all sounds more like a research group topic.

> >> That said, this assumption alone does not necessarily imply there is a
> >> state migration problem worth to be solved by the IETF. I am missing
> >> statements from VMware engineers (since you mention VMotion) or the
> >> like where they explain how IETF protocol work can help them to make
> >> their migration solutions work even better as they do.
> [Dacheng Zhang] 
> Agree. If we all agree that VM migration is happening or will happen in DC, it is the right time to discuss where there are any important scenarios where state migration across network side devices is necessary.
> 
> By the way, there is a Cisco white paper " Virtual Machine Mobility with Vmware VMotion and Cisco Data Center Interconnect Technologies". I personally think it is quite interesting. Anybody having interests can download it and have a look. ^_^
> 

If I understand the paper correctly, then they assume an extended VLAN
with a common IP subnet is in place for cross DC migration to work and
they the leave migration across IP network boundaries as something to
be worked out in the future.

/js

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