Re: patch for the checkpointing paper
Please note that as per our previous discussions, we aren't going
to be adding the 'cache' system call and you need to retool the
calls you are currently making to just pass a flag down to the
existing checkpoint/restore routines to tell them not to save/restore
the descriptor set, and that you have to use the normal file path
mechanism to create the files and not try to have the kernel do it
itself, and we are not going to store a state variable in the proc
structure, and a myrid of other issues which I have already discussed
with you. Most of these items should be handled strictly within userland,
using file-backed shared memory to cache persistent data that you do
not want to be restored and to track state. The patch set cannot not be
committed in its current form.
-Matt
討論串 (同標題文章)