RPC and distributed objects summary
-----------------------------------


RPC: goal is to simplify development of distributed applications

- abstracts away the details of communication
- make remote calls as simple as local ones

* on client side, user stub acts as a proxy for the remote procedure
* user stub "marshals" call parameters into a message, and uses the
  RPC runtime library to communicate with server
* server stub (skeleton) demarshals parameters and forwards request to
  remote procedure

naming mechanism is needed to allow clients to locate servers;
Grapevine filled this need in the original RPC system

implemented on an unreliable datagram protocol (like UDP)
advantage: less overhead for short, self-acknowledging
  request-response style communication
disadvantage: transferring large parameters is tricky


Lightweight RPC: optimize for the common case, i.e. intra-domain
communication

- let client threads execute requests in the server's space to avoid
  message-passing overhead
- pre-allocate shared buffers for call parameters and results
- cache domain contexts on idle processors (which must have been
  abundant in 1990)


Network Objects: object-oriented RPC

key advantage over RPC: ability to pass around remote object
references, allowing transparent access to state as well as to
functionality

key challenge: object-oriented data structures tend to be more complex
than standard ones, so marshaling parameters in a general way is a
challenge


CORBA: heterogeneous object-oriented RPC

emphasis on supporting multiple platforms, applications, network types
interfaces are expressed in CORBA IDL

supports transparency of:
- object location
- object implementation
- object execution state (running or not)
- object communication mechanisms

major flaws, corrected later on:
- no interoperability between ORBs
- can't pass objects by value


Information Bus: publish/subscribe in addition to RPC

goal: continuous operation while supporting dynamic evolution as well
as legacy systems

major application: managing constant flow of information among
loosely-coupled processes