If I recall correctly, they actually wacked the OS. They used /proc
and made a way that another process could map a different process'
address space. Sends can be implemented, on page sized quantities,
as mapping a page COW into another process' address space.
Zero copy for big data that wasn't touched, otherwise copy on write.
For small data, it was just a copy and a wakeup.
I may have the details wrong, but the bottom line was that MPI went
very, very fast - customers were quite amazed and very pleased. It
exceeded their expectations.
-
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/