Concurrency

1
Concurrency

• Outline potential applications of concurrency
• Explain typical problems with concurrency
• Use appropriate vocabulary

2
Example 1 Responsiveness

• Maintaining a responsive user interface
• While communicating
• While performing complex processing

3
Example 2 Multiple External Actors

• Bank
• Distributed system
• Simultaneous activity
• Components interact with body of information
• Must maintain consistency

4
Example 3 Efficiency

• Making use of idle time
• Two tasks both mixing processing I/O
• Sequential one after another
• Parallel using processor when awaiting I/O

Task 1
Task 2
Reduces total time average waiting time
5
Example 4 Power / Speed

• Parallel Systems
• Multiple processors tackling the same problem
• Pipelining
• Each processor carries out a step of the process
• Division of labour
• Each processor works on part of the data

6
Are Concurrent Programs Needed?

• The world is concurrent
• Programs must handle simultaneous activities
• Cant predict order of requests
• Cant write a sequential program
• Processor speed is limited
• Parallelism is only way to increase speed

7
Handling Concurrency

• Threads
• Simultaneously executing components
• Use one thread to manage an external actor
• Event Handlers
• Blocks of code triggered by an event
• Keep event queue handle events sequentially

Event
Event handler
Event handler
Event Queue
Event handler
8
Concurrent Components

• Other terms
• Activity
• Task
• Fiber
• Co-routine

• Application
• Complete program
• OS may run several
• Process
• Code Data State of execution
• Heavyweight
• Has own memory space
• Thread
• Code Data State of execution
• Lightweight
• Shares memory space with other threads

Processes
Thread-like, co-operatively scheduled
9
Parallel Concurrency?

• Parallel Systems
• Multiple processors working simultaneously
• Concurrent Systems
• Executing multiple programs at the same time
• May see progress on several programs by sampling

Time-slicing
Concurrent
Parallel
Multi-processor
10
Parallel ? Concurrency does it matter?

• No!
• Design all concurrent programs like parallel ones
• You cannot tell when the processor will switch
• Any thread can be interrupted part way through
• Even on a single processor system
• Otherwise, your program will have bugs
• And they will be difficult to find
• Intermittent
• Time-dependent
• Hard to reproduce

NO Treat all programs as parallel
11
Java Concurrency

• Threads built into the language
• Portability
• Synchronisation primitives built in to objects
• Each object has a lock
• Well-behaved threads seize lock before
  manipulating the object
• Ensures mutual exclusion
• Low-level can be misused
• No protection against deadlock
• Java 2 provides additional primitives

12
Java Threads

• A class with a run() method that
• extends (inherits from) Thread
• Activated by the start() method
• Creates a new thread,
• Executes run() in parallel
• Alternatively a class that
• implements Runnable (an interface)
• Provides a run() method
• Used to create a thread
• Thread myThread new Thread(myRunnable)
• myThread.start()

13
Example

• public class NumberThread extends Thread
• int num
• public NumberThread(int num) this.num
  num
• public static void main(String args)
• NumberThread numberThread new
  NumberThread5
• for(int i0 i lt 5 i)
• numberThreadi new
  NumberThread(i)
• numberThreadi.start()
• static void display(int val)
• for (int rows 0 rows lt 4 rows)
• for (int i 0 i lt 35 i)
• System.out.print(val)
• System.out.println()

14
Unsynchronized Output

• Interleaving of output
• Results vary at random
• Need mutually exclusive access to screen

15
Locks

• The first task to seize a lock can proceed
• Later tasks seeking the lock are suspended
• Until the lock is released

16
Example

• public class NumberThread extends Thread
• int num
• public NumberThread(int num) this.num
  num
• public static void main(String args)
• NumberThread numberThread new
  NumberThread5
• for(int i0 i lt 5 i)
• numberThreadi new
  NumberThread(i)
• numberThreadi.start()
• synchronized static void display(int val)
• for (int rows 0 rows lt 4 rows)
• for (int i 0 i lt 35 i)
• System.out.print(val)
• System.out.println()

Only one thread at a time can execute this
method
17
Synchronized Output
18
Priority Inversion Mars Pathfinder

• Pathfinder had an "information bus,
• a shared memory area for data exchange
• Bus access synchronized by a lock
• mutual exclusion lock (mutex).
• Threads have priorities reflecting their urgency.
• There were occasional problems
• Little green gremlins?

19
Priority Inversion Mars Pathfinder

• A low priority weather thread ran infrequently
• acquire mutex, write to the bus, release mutex.
• If the high priority bus management thread ran
  while the mutex was held, it blocked, waiting for
  the weather thread to release the mutex
• If a medium priority communications thread was
  scheduled during this short time, it prevented
  the weather task running, blocking the management
  task.
• A watchdog timer would trigger, notice the data
  bus task had not run recently, decide there was a
  problem, reset the system, losing data.
• Fix give threads holding a mutex the maximum
  priority of waiting threads

Information Bus
Low
mutex
High
Low
Medium
20
Vocabulary

• Critical Region
• Sections of code which could cause problems if
  multiple threads simultaneously execute them
• Synchronisation
• Co-ordinating threads processes
• Examples of synchronisation
• Mutual Exclusion
• Ensuring at most one thread is in a critical
  region
• Communication
• Passing data between threads

21
Summary

• Concurrency is essential
• Treat concurrent activities as if parallel
• Java provides threads
• Treat concurrent activities as if parallel
• Threads / Processes
• Concurrent program parts
• Threads share memory space
• Share global variables
• Consistency problems unless care taken
• Treat concurrent activities as if parallel