RPC and client-server systems
e.g. File System
RPC: usually remote
client-server on same machine?
-performance vs. safety
RPC vs. Simple Procedure call
Procedure call
caller callee
process: all at compile time
RPC call – return
call trap in the Kernel – server procedure
return trap
All at run time
copying overhead
client stack -> client stub -> rpc msg -> kernel -> kernel buffer -> server -> server domain -> server stub -> server stack
How to remove overheads?
-set up (binding) => one time cost
Entry point, A-stack size, call
c s.foo import s foo => name server – grant kernel PD
client adr space = sm a-stack = server adr space
Making RPC cheap(actual calls)
client address space -> args -> A-stack -> result
Kernel
-> server address space
Original client stack
-> rpc msg -> kernel buffer -> server domain -> server stack
Now
client stack -> A-stack shared -> server stack : two copies, marshal, unmarshal
