The Slice prototype uses a Trapeze-based kernel-kernel RPC.
- derived from the original gms_net module for Trapeze
- handles block I/O traffic as attached payloads
unified buffering for zero-copy block traffic
- supports asynchronous RPCs, e.g., for prefetching
calling process supplies a procedure to execute on reply
- optimized to reduce context switching overhead
Optimistically handle RPC replies at interrupt time, with a primitive to defer work to a process context.