[Planetlab-devel] Re: [Planetlab-arch] planetlab - slice migration a possibility?

[Swetha Krishnan]

Sorry that we were stuck with some other deadlines, and so started on 
our sliver migration venture only recently. We did get PLC to start up, 
and are working on myplc-devel now.

We were wondering if there is some documentation for the code itself - 
is there something that describes the structure and organization of the 
code, that would help us read the code in order from a starting point, 
and according to relevance for our project?

We looked at /build/2007.03.21/SOURCES, but there isn't a README there. 
(although there is some stuff in the sub-directories such as util-vserver)

Would really appreciate any help.

Thanks,
Shweta and Giri.


Marc E. Fiuczynski wrote:
> Hello Swetha,
> 
> This would be a neat feature.  There are folks working on bringing 
> container-based OS technology (e.g. Linux-VServer) to Linux mainline. 
> These same folks care about "container migration", which basically is 
> process migration done right, to Linux.  What's neat about container 
> migration is that folks have demonstrated that they can migrate from one 
> version of the Linux kernel to another (i.e., from 2.6.16 to 2.6.18).
> 
> While Linux-VServer does not have this functionality, purportedly (i.e., 
> I have not tried myself) the OpenVZ project does support this feature 
> today and the opensource solution is available right now.  There's also 
> a concept implementation from IBM called MCR available on sf.net.  You 
> might want to look into OpenVZ, join the containers mailing list @ osdl, 
> and generally get plugged into that community in order to have impact.
> 
> PlanetLab might pick this up when the said technology makes it ways into 
> mainline Linux within the next 12 months or so.  The main difficulty 
> with migrating a container on a system like PlanetLab is that the normal 
> assumption is that the unshared IP address migrates with the container 
> (i.e., sliver using PL parlance).  This wont be the case for IPv4 
> address, as PL generally only has one of those assigned per host.  It 
> would be good to look into how well IPv6 works with this functionality, 
> especially leveraging mobile IPv6 to ensure the best communication 
> routes are taken for a container that has been migrated from a machine 
> at one site to another machine at another site.
> 
> Hope this helps.
> 
> Best regards,
> Marc
> 
> 
> Swetha Krishnan wrote:
>>
>> I am a newbie to PlanetLab and have just read the papers (never really 
>> used planetlab before). I am interested in doing a project for adding 
>> a feature to planetlab - live migration of slices.
>> I am not precisely sure if this is already something that's taken care 
>> of in planetlab, but imagine this: a user (i.e. user experiment) who 
>> is say allocated x slices on host A, y on host B and z on host C. Now 
>> if host B 's utilization bumps up due to some reason, or its memory is 
>> running scarce, could we dynamically move some or all of those 'y' 
>> slices to one of the other hosts, instead of having to kill these 
>> slices? It is such dynamic migration support that I wish to add.
>> If this sounds a project that's a possibility, could you let me know 
>> whether the linux vserver mechanism that planetlab uses, has any 
>> existing support for process migration (like Xen has)..?
>>
>> Thanks,
>> Swetha.
>>
>> _______________________________________________
>> Arch mailing list
>> Arch at lists.planet-lab.org
>> https://lists.planet-lab.org/mailman/listinfo/arch
>