|
|
Ok,
So I have resubmitted the ckpt patch, along with a little bugfix (it did
just checkpoint the first 30 seconds... it never restarted the timer after
that).
I've been playing with GNU Pth, their userspace pthread implementation.
While it worked ok (this is not incredible due to blocking I/O) without
checkpointing, it actually makes the checkpointing process segfault... which
means that going userspace didn't fix anything. This is why Pth patches are
not being submitted for now (anyways, that was 2 line...).
So we have a problem, and I believe the problem is the ckpt is not made to
checkpoint threads. So, I still need to find a solution for checkpointing...
Because I think I did what I could on the thread side... Pth seems to be
good enough for our needs if we can handle the I/O stuff properly.
Jacques Mony
|
| |
