Caps status
Matthew Dillon
dillon at apollo.backplane.com
Wed Jan 21 01:28:32 PST 2004
:Cool... sounds like out of band mach messaging in a way. Except that
:I think it does full COW for certain sized messages. Its more like
:a shadow copy until some modification is done which of course means
:the pages are shared and the message reception "appears" very fast.
Well, the absolute fastest way is to maintain some persistent shared
memory between client and server and store the messages and replies
inline. Any time you have to continually map and unmap memory you are
not going to have good performance (and the current caps interface has
the same issue with its copyin's and copyout's).
Shared memory isn't a bad way to go but it does create restrictions on
the message size. We could in fact support a persistent shared memory
scheme by adding a virtual pointer feature where the client and server
negotiate a shared memory segment. Then just the data pointers
are passed back and forth in the messages and not the data itself.
-Matt
More information about the Kernel
mailing list