• Dates: 2/6-11/03:
• Reading: Ch 7.

General problem

• independent vs. cooperating processes
• earlier examples: reader/writer, consumer/producer (bounded buffer), etc.
• general assumption: shared resources
  typically memory/data variables
• potential problems among concurrent processes
  • consumer/producer example in book
  • item count update by both consumer and producer
  • multiprogramming: interleaving condition
  • truly parallel: parallel access
  • race condition: execution results dependent on exact order
• general situation: (data) dependency analysis
  • read/write (or define/use)
  • relations: DD, DU, UD, UU
  • followup analysis in testing course (CSE 7314)
• critical-section: a general/important problem
• other problems: process cooperation/IPC/synchronization

Critical-section problem

• critical section (CS): limited access
• narrow sense: at most 1 access
  also called mutual exclusion problem
• solution requirement:
  • mutual exclusion
  • progress
  • bounded waiting (no starvation)
• general setup: repeat { S1; CR; S2 }
• solution: entry and exit conditions
• solution strategy: software vs. hardware

CS problem: Software solutions

• two process software solution
  then general solution

• algorithm 1: (Fig 7.2, p.193)
  • use of turn
  • entry: check turn
  • exit: set turn to other
  • problem: progress when not turn

• algorithm 2: (Fig 7.3, p.194)
  • flags
  • entry: set own flag and see other flag
  • exit: reset own flag
  • problem: progress when both raise flags
    (similar to deadlock problem)
  • change entry order: mutex, both enter

• algorithm 3: (Fig 7.4, p.195)
  • both turn and flag
  • entry: raise flag, give other turn,
    yield if other raised flag and have turn
  • exit: reset flag
  • why does it work?
  • flag: signal
  • turn: progress when contention occurs

• multiprocess solution:
  bakery algorithm, Fig. 7.5, p.197
• difficulty of software solutions

CS problem: Hardware solutions

• single CPU: disabling interrupt
• atomic operations: test-and-set or swap

• "test-and-set" solution:
  • definition: Fig 7.6, p.198
  • solution: Fig 7.7, p.198
  • global exclusive "lock" or "token"
  • whoever performed it (atomically) first, will set & proceed
  • whoever performed it later, will wait
  • un-set (lock = false) initially
  • un-set when done

• "swap" solution:
  • definition: Fig 7.8, p.199
  • solution: Fig 7.9, p.199
  • similar to above
  • global "lock"
  • local "key"

• problem of bounded waiting?
• general solution Fig 7.10, p.200
  read and understand yourself

CS/other problems: Generalized solutions with semaphores and monitors

• semaphore: generalized concept/tool
• concept: integer variable
• access only through P and V
  P = wait, might cause blocking
  V = signal, increment
  classical definition p.201
• P, V, indivisible (atomic) operations
  on value change
  between "while" and "S--" in P.
• use in critical section problems
  P(S) CS V(S), with S initialized to 1
  critical section problem for any number of processes
• implementation:
  blocking (p.203) instead of
  busy waiting (p.201) or use spinlock
• solution to general IPC/cooperating process problem
• limited access instead of exclusive access section
• multiple semaphores: possibility for deadlock
• binary vs general semaphores

Other classical coordination problem

• problems
  • consumer-producer or bounded buffer
  • reader-writer
  • dining philosopher
• solution use semaphores
• general ideas (class), and then code (textbook)
• also, time for review

• consumer-producer or bounded buffer
  • buffer size
  • problems with update
  • waiting conditions:
    • full: producer
    • empty: consumer
    • both: update item count
  • use of three semaphores
  • solution: Fig 7.12 & 7.13, p.207

• reader-writer
  • asymmetric processes: two types
  • reader: multiple access possible
  • writer: exclusive access
  • waiting/blocking conditions :
    • writer writing => everyone waiting
    • reader reading => writer waiting
  • key solution idea:
    • reader as a "block" or "group"
      first to enter or last one to leave need to act for the whole group
    • writer as individuals
  • solution for
    writer (Fig 7.14, p.208)
    reader (Fig 7.15, p.209)

• dining philosopher
  • problem description
  • key difference: distributed vs. centralized environment as before
  • limited resource sharing
  • communication through message passing
  • solution
  • deadlock possibility
  • starvation possibility
  • central knowledge implications

Review for exam#1

• question types:
  • mostly similar to homework questions
  • multiple choice also
  • some homework problems
• key material:
  • processes and threads; ideas & concepts
  • scheduling: ideas and algorithms;
  • synchronization/coordination:
    concepts and CR/classical problem solutions
  • general ideas about chapters 1-3
• 3 or 4 question sets (3-4 pages long)

Material to cover after exam#1

• monitors
• concept: generalized semaphores
• .wait and .signal on different conditions
  critical regions (or conditional cr) problem
• use in general IPC problems