Proportional share scheduler
dillon at apollo.backplane.com
Thu Jul 3 10:28:11 PDT 2008
:Well, migration could be partly implemented using process checkpointing.
:I guess what is missing in DragonFly's process checkpointing is
:remembering the state of the associated files/sockets. Thinking about
:it, in the case of a vkernel, this should be no big deal. Anyone ever
:tried to checkpoint a running vkernel?
:That would be pretty cool to just checkpoint a running vkernel and
:reload it on any other machine (with a somewhat similar configuration).
:But I'm really missing a use case for myself right now ;-)
It is possible to do this. The vkernel only has a few open file
descriptors... its raw disks and network interface and that's pretty
much it. The only other things that would need to be dealt with are
the cpu and I/O threads and the special virtual page table memory
mappings. Both are easy to reconstruct.
For the real kernel we have the beginnings of a universal identifier
(that's what all the sysref stuff is), and ultimately the idea is to
allow resources to be migrated piecemeal while the processes are
<dillon at backplane.com>
More information about the Kernel