Distributed memory had early vogue pre-RPC, and is coming
back on very reliable fast local networks. It isn't clear how
it integrates with most safe modern programming techniques,
but as a low level structuring tool in low latency
situations, it may be very nice.
-
User level process locks at memory reference cost
-
Flexible memory segment control between groups of processes
-
Low cost system call interface ;SPM_gt;14uS (~250 instructions)
on our SGI MP machines
-
A fork which allows selective sharing of various resources including
memory, file tables, locks, environment and notes.
-
Two synchronisation system calls to interface to the scheduler
-
rendezvous - to provide a cheap synchronisation primitive between
processes in the same process group.
-
sleep(0) - to reschedule the processor