Re: [RFC] Container mini-summit agenda for Sept 3, 2007

[Oren Laadan <orenl-eQaUEPhvms7ENvBUuze7eA@public.gmane.org>]

Cedric Le Goater wrote:
> Hello All,
> 
> Some of us will meet next week for the first mini-summit on containers.
> Many thanks to Alasdair Kergon and LCE for the help they provided in 
> making this mini-summit happen !
> 
> It will be help on Monday the 3rd of September from 9:00 to 12:45 at LCE
> in room D. We also might get a phone line for external participants and, 
> if not, we should be able to set up a skype phone.
> 
> Here's a first try for the Agenda. 
> 
> Global items 
> 
> 	 [ let's try to defer discussion after presentation ]  
> 
> * Pavel Emelianov status update 
> * Serge E. Hallyn Container Roadmap including  
> 	. task containers (Paul Menage)
> 	. resource management (Srivatsa Vaddagiri) 
> 
> Special items
> 
> 	[ brainstorm sessions which we would like to focus on ]
> 
> * builing the global container object ('a la' openvz or vserver)
> * container user space tools
> * container checkpoint/restart  

        5. checkpoint/restart
                memory c/r
                        (there are a few designs and prototypes)
                        (though this may be ironed out by then)
                        per-container swapfile?
                overall checkpoint strategy  (one of:)
                        in-kernel
                        userspace-driven
                        hybrid
                overall restart strategy
                use freezer API
                use suspend-to-disk?

                sysvipc
                        "set identifier" syscall
		pid namespace
                        clone_with_pid()
There are other identifiers - pseudo terminals, message queues (mq)
(if you insist on supporting these ...). In general, we need a way
to specify the virtual id of a resource that is created. I suggest
that this should be part of an interface between c/r and containers
(see below)

		live migration
aka pre-copy (which can be used for live migration but also to reduce
the downtime due to a checkpoint).

how about adding incremental checkpoint to the list ?

I think that it is also important to discuss an interface between c/r and
containers, each of which stands on it own. For instance, how to request
a specific virtual id (during restart), define required notifiers (to
set/unset c/r related data on/off a task), control c/r-related setting of
container (e.g. frozen, restarting) that may affect behavior, such as
signal handling, and so forth. Also, such an interface can allow existing
c/r implementations to work with different virtualization implementations
as they become available.

Many of these were discussed in a recent Zap paper present in USENIX:
http://www.ncl.cs.columbia.edu/publications/usenix2007_fordist.pdf
The paper describes important design choices in Zap (but I'm biased ...).
I think it may serve as an appetizer for the discussion :P

Oren.