1. OS Fall’02 Concurrency: Principles of Deadlock Operating Systems Fall 2002

2. OS Fall’02 Processes and resources  Processes need resources to run CPU, memory, disk, etc … A process waiting for a resource cannot complete its execution until the resource becomes available  There is only a finite amount of resources E.g., 1 CPU, 1 GB memory, 2 disks

3. OS Fall’02 Concurrency and deadlocks  In a multiprogramming system the total resource demand by all concurrently active processes exceeds by far the total amount of available resources  Processes compete for resources A process can grab the last instance of a resource A and wait for the resource B Another process may hold B and wait for A No one can proceed: Deadlock

4. OS Fall’02 Deadlock  Permanent blocking of a set of processes that either compete for system resources or communicate with each other  Involves conflicting needs for resources by two or more processes  There is no satisfactory solution in the general case

5. OS Fall’02 Deadlock in the everyday life 1 23 4

6. OS Fall’02 Deadlock in the everyday life

7. OS Fall’02 Deadlock when contending for the critical section

8. OS Fall’02 Example of Deadlock Progress of Q Release A Release B Get A Get B Get AGet BRelease ARelease B Progress of P A Required B Required A Required B Required deadlock inevitable 1 2 3 4 5 6 P and Q want A P and Q want B

9. OS Fall’02 Example of No Deadlock Progress of Q Release A Release B Get A Get B Get ARelease AGet BRelease B Progress of P A Required B Required A Required B Required 1 2 3 4 5 6 P and Q want A P and Q want B

10. OS Fall’02 Resource categories  Reusable: Used by one process at a time and not depleted by that use  can be reused by other processes,may exist several instances Processors, memory, disks, tapes, etc.  Consumable: Created (produced) and destroyed (consumed) by a process  Interrupts, signals, messages, and information in I/O buffers

11. OS Fall’02 Reusable resources and Deadlock  Deadlock might occur if each process holds one resource and requests the other  E.g., Space is available for allocation of 200K P1... Request 80K bytes; Request 60K bytes; P2... Request 70K bytes; Request 80K bytes;

12. OS Fall’02 Consumable resources and Deadlock  Example: Deadlock occurs if receive is blocking P1... Receive(P2); Send(P2); P2... Receive(P1); Send(P1);

13. OS Fall’02 Conditions for Deadlock  Policy conditions Mutual exclusion Hold-and-wait No preemption  Circular wait Resource B Resource A Process P1 Process P2 Requests Held by Requests Held By

14. OS Fall’02 Conditions for Deadlock Mutual exclusion Hold-and-wait No preemption Circular wait DEADLOCK

15. OS Fall’02 Circular Wait P1P2P3 R1 R2 R3

16. OS Fall’02 No circular wait P1P2P3 R1 R2 R3

17. OS Fall’02 Coping with Deadlocks  Deadlock prevention Deadlock possibility is excluded a priori by the system design  Deadlock avoidance Deadlocks are possible in principle but avoided  Deadlock detection Deadlocks can occur: detect and solve the problem

18. OS Fall’02 Deadlock prevention  Design system so that it violates one of the four necessary conditions Prevent hold and wait:  request all the resources at the outset  wait until all the resources are available Prevent circular wait by defining linear ordering of the resource types  A process holding some resources can request only resource types with higher numbers

19. OS Fall’02 Preventing circular wait P1P2P3 R1 R2 R3

20. OS Fall’02 Deadlock prevention: Cons  Degraded performance Delayed execution Low parallelism  Hold and wait prevention is wasteful Hold resources more than they are needed When might this be reasonable?

21. OS Fall’02 Deadlock avoidance  Allocate resources in a way that assures that the deadlock point is never reached  The allocation decision is made dynamically based on total amount of resources available currently available processes ’ resource claim processes ’ current resources allocation

22. OS Fall’02 Banker ’ s algorithm ( Dijkstra 65 ’ )  Do not grant an incremental resource request to a process is this allocation might lead to deadlock  The system state: is the current allocation of resources to processes  Safe state: is a state in which there is at least one sequence in which all processes can be run to completion  Unsafe state = NOT safe state

23. OS Fall’02 Determination of the safe state  We have 3 resources types with amount: R(1) = 9, R(2) = 3, R(3) = 6  Is the state S0 below safe? Claim Allocated Total 3 2 2 6 1 3 3 1 4 4 2 2 1 0 0 6 1 2 2 1 1 0 0 2 P1 P2 P3 P4 R1 R2 R3 9 3 6 Available 0 1 1

24. OS Fall’02 Determination of the safe state Claim Allocated Total 3 2 2 0 0 0 3 1 4 4 2 2 1 0 0 0 0 0 2 1 1 0 0 2 P1 P2 P3 P4 R1 R2 R3 9 3 6 Available 6 2 3 Claim Allocated Total 0 0 0 3 1 4 4 2 2 0 0 0 2 1 1 0 0 2 P1 P2 P3 P4 R1 R2 R3 9 3 6 Available 7 2 3

25. OS Fall’02 Determination of the safe state Claim Allocated Total 0 0 0 4 2 2 0 0 0 0 0 2 P1 P2 P3 P4 R1 R2 R3 9 3 6 Available 9 3 4 Claim Allocated Total 0 0 0 P1 P2 P3 P4 R1 R2 R3 9 3 6 Available 9 3 6 S0 is safe: P2->P1->P3->P4

26. OS Fall’02 Banker ’ s algorithm  When a process request resources: Assume the request is granted Update the system state accordingly Determine whether the resulting state is safe If yes: grant the resources Otherwise, block the process until it is safe to grant the resources

27. OS Fall’02 Banker ’ s algorithm Claim Allocated Total 3 2 2 6 1 3 3 1 4 4 2 2 1 0 0 5 1 1 2 1 1 0 0 2 P1 P2 P3 P4 R1 R2 R3 9 3 6 Available 1 1 2 P2 requests (1, 0, 1): Grant or not? P1 requests (1, 0, 1): Grant or not?

28. OS Fall’02 Deadlock detection  Banker ’ s algorithm is Pessimistic: always assume that a process will not release the resources until it got ’ m all  decreased parallelism Involves complicated checks for each resource allocation request (O(n^2))  Optimistic approach: don ’ t do any checks When deadlock occurs - detect and recover Detection: look for circular waits

29. OS Fall’02 Practice  Most operating systems employ an “ ostrich ” algorithm  Break hold-and-wait Cannot acquire a resource - fail back to the user:  e.g., too many processes, too many open files  Quotas  Programming discipline: acquire locks (semaphores) in a specific order

30. OS Fall’02 Dining philosophers problem

31. OS Fall’02 Dining philosophers problem  An abstract problem demonstrating some fundamental limitations of the deadlock- free synchronization  There is no symmetric solution  Solutions execute different code for odd/even give ’ m another fork allow at most 4 philosophers at the table Randomized (Lehmann-Rabin)

32. OS Fall’02 Concurrency: summary  Critical section is an abstract problem for studying concurrency and synchronization software solutions hardware primitives higher level primitives: semaphores, monitors  Deadlocks are inherent to concurrency 4 conditions 3 ways to cope with

33. OS Fall’02 Next: Memory management