Remote Procedure Call - Operating Systems | CMSC 412, Study notes of Operating Systems

Download Remote Procedure Call - Operating Systems | CMSC 412 and more Study notes Operating Systems in PDF only on Docsity!

copyright 1996 Jeffrey K. Hollingsworth

Announcements

l^

Reading Chapter 17 (skip 17.6.1 and 17.6.4)– problems: 17.1, 17.3, 17.

copyright 1996 Jeffrey K. Hollingsworth

Remote Procedure Calls

l^

Provide a way to access remotes services

l^

Look like “normal” procedure calls

l^

Issues:– binding functions to services

• can use static binding (like kernel trap #’s)• can use a nameserver
  • data format
    • different machine may have different formats• translation is called

marshalling

• pick a common way to encode info (e.g. XDR)– always send in this common format
  • failures
    • what if a host dies while and RPC is active?

copyright 1996 Jeffrey K. Hollingsworth

RPC Generators

l^

Given a list of functions to make into RPC

l^

Generate the code for:– RPC stubs (for clients to call)

• marshalling code for each parameter• utility routines to marshal structures/records• code to send messages and wait for responces
  • Server code
    • case statement for each RPC type• un-marshal parameters• call local routine
      • detecting errors– checking version numbers between client/server

copyright 1996 Jeffrey K. Hollingsworth

Failures

l^

Fail Stop– system either produces the correct answer or no answer– hard to know “what” failed

• local network card• network link• remote network card• remote system• remote software

l^

Byzantine Failure– systems can “lie” and produce wrong answers

• a message shows up but some of the data is wrong
  • can use check sums to detect this failure mode
    • does not deal with malicious failure
      • considered a “hard” problem

copyright 1996 Jeffrey K. Hollingsworth

DFS Performance Issues

l^

“normal” filesystem issues– disk parameters: seeks time, rotational latency– filesystem time: directory structure, fat/inodes

l^

distributed system issues– network:

• latency (time for small requests)• bandwidth (time to move entire disk blocks)
  • coordination
    • time to access servers
      • namespace server– fileserver

copyright 1996 Jeffrey K. Hollingsworth

Caching

l^

To improve performance, cache DFS information– goal: improve response times for overall DFS

l^

Local Cache– memory cache

• data is stored in memory of local system
  • disk cache
    • data is stored on the disk of the local system

l^

Server Cache– memory

• can put lots of memory here so most “popular” files are in memory