Communication in Distribute System

#Engineering

#Distribute

This code is called inside callout

Issue in Client-Server communication

Addressing

Hard-wired address: Machine address and process address are known
Broadcast-base:
- server chooses address from a sparese address space
- Client broadcasts request
- Can cache response for future
Using Nameserver

Blocking vs Non-blocking

Blocking communication is synchronous means:
- Send blocks until message is actually sent
- Receive blocks until message is actually received
Non-blocking communication is asynchronous:
- Send return immediately
- return does not block either

Buffered vs Unbuffered

Unbuffered communication: server must call receive before client send
Buffered communication:
- Client send to a mailbox
- Server receives from a mailbox

Reliable vs Unreliable

Unreliable channel
- Need acknowledgements (ACKs)
- Application handle ACKs
- ACKs for both request and reply
Reliable channel
- Reply acts as ACK for request
- Explicit ACK for repsonse
Reliable communication on unreliable channels
- Transport protocol handles lost messages

Remote Procedure Calls (RPC)

What is RPC ?

Goal: Make distributed computing look like centralized computing
Allow remote services to be called as procedures
- Regardless of location, implemention, language
Issues:
- How to pass parameters
- Bindings
- Sementics in face of errors
Two classses: Integrated into prog, language and separate

Parameter passing

Local procefure passing by "callByValue" or "Call by reference". RPC simulate this through

Stubs - Proxies: Take care of packaging arguments and sending messages, stub compiler generates stubautomatically from specs in an IDL (Interface Definition Language)
Flattening - Marshalling: packaging parameters

Steps

Client procedure calls client stub
CLient Stubs builds message, calls local OS
client's OS sends message to Server's OS
Server's OS give message to server Stub
Server Stub unpacks parameters, call server
Server does work, return result to stub
Repeat the steps to send message to Client

Marshalling

Standard Representation (Ex: XDR, ...)
Passing pointer
- If point to well defined data structure, pass a copy and stub poasses pointer to local copy
- Prohibit data structures containing pointers
Transform parameters/results into a byte stream

Binding

Server: export server interface during initialization. Also, send name, version, unique identifier, handle (address) to binder Client: First RPC send message to binder to import server interface. Binder check server exported interface and return [handle and unique identifier] to client

RPC models

Lightweight RPCs: optimize for communicate between client and server but on same machine
- Push arguments onto stack
- Trap to kernel
- Kernel chanes mem map of client to server address space
- Client thread executes procedure (OS upcall)
- Thread traps to kernel upon completion
- Kernel changes the address space back and returns control to client
Asynchronous RPC
Deferred-synchronous RPC
One-way RPC

RPC vs REST Api

Similarities

Both are API design approaches, allow two distinc services communicate without knowning detail knowledge about each others allow two distinc services communicate without detail knowledge about each others. + Abstraction: abstract lower level communications in network communications. + Communication medium: both use HTTP under the hood with JSON/XML,... + Multi-language Compatibility

Differences

RPC focus more on functions or actions while REST focus on resources or objects (CRUD)
RPC usually used for calling complicate actions/functions in remote server while REST usually used for action on objects (CRUD)
RPC can be statefull or stateless while REST API must be stateless
RPC is suitable for complicated task on remote services, REST is suitable when both client and server need to be uniform represented
RPC parameters structure is defined by remote services while REST allow sending alternatives structure (JSON, XML, ...) clients have freedom to choose the data type they want