Remote Procedure Call Andy Wang Operating Systems COP

Remote Procedure Call Andy Wang Operating Systems COP 4610 / CGS 5765

Primitives to Build Distributed Applications n send and receive Used to synchronize cooperating processes running on different machines n Communicate through mailboxes (ports) n n n Temporary holding areas for messages Atomic operations n send/receive the entire message or nothing

More on send and receive n To put a message to the network n send(message, destination_mbox) n receive(buffer, mbox) n n Waits until a message arrives Copies the message into a given buffer

Remote Procedure Call (RPC) n Allows you to invoke a procedure on either a local or remote machine n Client n n Remote. Machine Say(“Meow”); Server n Machine Say(“Meow”); n Implemented on top of two-way messaging

RPC Illustrated Client (caller) call Server (callee) reply Client stub call Server stub OS or Network

Procedure Stubs n Provide the invocation interface programmers n e. g. foo(int a) n Client stub n Build messages (a. k. a. marshalling) n Send messages n Wait for response n Unpack reply n Return result

Server Stub n Create N threads to wait for requests n Loop n Wait for command n Decode and unpack request parameters (unmarshalling) n Call procedure n Build replay message with results n Send reply

RPC vs. Procedure Call n From the programmer’s viewpoint n Similar semantics n Pointers are instantiated before transmission n Data structures pointed by the pointer are copied n Processes running on the remote machine is in a different address space

Implementation Issues n Stubs are automatically generated n Need to have a well-known port to talk to servers n Server can upgrade the implementation without recompiling client applications

Interprocess Communication n RPC is just another way to communicate between processes n Example uses n Microkernel operating systems n n Portions of an OS are implemented at the user level to minimize the kernel-level code Object linking and embedding (OLE) n Mix-and-match applications

Using RPC for IPC + Fault isolation: bugs are unlikely to propagate across different address spaces + Modularity: independent component upgrades + Location transparency: a service can be provided from local or remote machines

Using RPC for IPC - Poor performance - A wide range of failure modes A user-level bug can cause a process failure n A kernel-level bug can cause multiple processes to fail n A network failure, server failure, or an earthquake can cause multiple machines to fail n