No problem, I don't really want to go into GNU/Linux versus Linus
Kernel either.
> Yes, but we're talking about checkpoint migration of CPU and I/O
> intensive jobs to boxes that have free CPU resources.
Yes, but ... you asked for an API. That is the documented API for
filtering system calls. You can do 'proxy filtering' for most
system calls, and I have a lot of details on the ones that cause
problems, like "exit" and "mmap".
If it's not fast enough, you can consider speeding up the implementation.
You can also consider interface enhancements. The documented interface
to ptrace is very old and very slow (one word per system call).
Also remember that any system call filtering, including ptrace,
has zero cost for "CPU intensive" user-space work. If you are compiling
or ray-tracing or cracking codes, you do a lot of work per system call
issued. If you want your 'cp -a' command to migrate around the cluster
it's going to slow down.
Anyways, I am not here to suggest that you checkpoint any certain way,
I am just here to throw an answer to your question so you can add the
answer into your knowledge base.
Other good things to know: you can access the child's memory fast by
reading /proc/$pid/mem. I don't know if you can write it reliably
(Andi Kleen wrote the code into 2.2 but you will be the first soldier
on the beach if you use it).
Michael
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/