CSE 480 Lecture Notes
February 29, 1996

Mutual Exclusion

- critical section problem: disable interrupts, semaphores, critical regions, monitors

+ A Centralize Algorithm (Fig. 3-8)

Coordinator
  loop
    recv msg
    case msg of
      REQUEST: if nobody in CS then reply GRANTED
               else queue the REQ; reply DENIED
      RELEASE: if queue not empty then 
                  remove 1st on the queue
                  reply "GRANT"
    end case
  end loop

Client
  send(REQUEST)
  recv(msg)
  if msg = GRANTED then enter CS
  else recv(msg); enter CS
  send(RELEASE)

- guarantees mutual exclusion
- fair (FCFS)
- a single point of failure (Coordinator)
- if not explicit DENIED msg, then cannot distinguish permission denied
from a dead coordinator

+ A Distributed Algorithm

- Lamport Algorithm, Ricart-Agrawala Algorithm
- assume total ordering on msgs
- msgs are reliable (e.g., every msg acked)
- when a process wants to enter a CS, it builds a msg containing
the name of CS, process id, current time

Algorithm (handout, pp. 128-129 [SiSh94])

Examples: Fig. 3-9, Fig. 6-7,..,6-10

Theorem.  The Ricart-Agrawala Algorithm achieves mutual exclusion.
Proof: By contradiction.  Suppose two sites Si and Sj are in CS concurrently.
WLOG, assume that Si has earlier timestamp than Sj.
That is, Si received Sj's request after Si made its own request.
Thus, Sj can concurrently execute the CS with Si only if Si returns a REPLY
to Sj before Si exits the CS.  But, this is impossible since Sj has
later timestamp than Si.

- 2(N-1) msgs: (N-1) request and (N-1) reply msgs
- n points of failure (i.e., each process becomes point of failure)
can use explicit ack and timeout to detect failured processes
- each process need to maintain group membership; non-trivial for large
and/or dynamically changing memberships
- n bottlenecks since all processes involved in all decisions
- may use majority votes to improve the performance

+ A Token Ring Algorithm

- Fig. 3-10
- if the token is lost, it needs to be regenerated.
- detection is difficult since there is not bound on how long a process should
wait for the token.
- if a process can fail, it needs to be detected and by-passed.