checkpoint/restart milestone 1

Matthew Dillon dillon at apollo.backplane.com
Mon Sep 15 01:12:26 PDT 2003


:	Ah, sounds good!  I looked into this idea some time ago, when
:	you can suspend a process, store it's state in an image file
:	and then you can use it whenever you want for reloading it
:	back.
:
:	The only issue I had with such kind of thing was to do with the
:	state that the kernel keeps, like file descriptor tables, and
:	vnode references, which are invalidated after a reboot, because
:	a new table is created.
:
:	One way to solve this problem was to index into the filedesc
:	table and make out which files were open, save this information
:	in some sort part of the process image, and then when the
:	image is loaded, we recreate references to the files originally
:	referenced; although, this is just an overall perspective on the
:	topic.  I am sure there were other roadblocks which I faced.
:
:	Anyway, this sounds ilke a really good idea!  :-)
:
:	Regards,
:
:-- 
:Hiten Pandya
:hmp at xxxxxxxxxxxxx

    Yah.  Obviously the feature will start out being rather primitive, but
    it should be possible to add functionality incrementally without having
    to ripup the stuff already done.  I can envision (in the future) a 
    program setting up a SIGCHKPT signal, for example, which would allow the
    program to implement its own checkpoint restore function based on a
    resource list given to it by the system.

					-Matt
					Matthew Dillon 
					<dillon at xxxxxxxxxxxxx>





More information about the Kernel mailing list