Resource-Allocation Graph [111]

• no cycle IMPLIES no deadlock

• deadlock IMPLIES cycle (necessary condition)

• cycle IMPLIES maybe deadlock (but not sufficient condition)

• single instance resource AND cycle IMPLIES deadlock

• (necessary and sufficient)

Deadlock: Multiple Instance Resources [112]

DEADLOCK:

NO DEADLOCK:

Methods for Handling Deadlock [113]

• never let deadlock occur

• prevention: break one of the 4 conditions

• avoidance: resources give advance notice of maximum use

• let deadlock occur and do something about it

• detection: search for cycles periodically

• recovery: preempt processes or resources

• don't worry about it (UNIX and other OS)

• cheap: just reboot (it happens rarely)

Deadlock: Prevention [114]

• break mutual exclusion:

• read-only files are shareable

• but some resources are intrinsically nonshareable (printers)

• break hold and wait:

• request all resources in advance

• request (tape, disk, printer)

• release all resources before requesting new batch

• request (tape,disk), release (tape,disk), request (disk,printer)

• disadvantages: low resource utilization, starvation

Deadlock: Prevention [115]

• break no preemption:

• process 1 requests resources already allocated to process 2:

• process 1 forfeits its current resources

• if process 2 is waiting for other resources: process 2 forfeits

• used for resources whose state is easily saved/restored

• CPU registers and memory space

• not printers or tape drives

• break circular wait:

• order all resources by unique numbers (tape drives, etc.)

• processes request resources in increasing order

Deadlock: Avoidance [116]

• processes give advance notice about maximum usage of resources

• processes make actual requests when they need a resource

• avoidance algorithm: allocate request only if it yields a safe state

• a sequence of processes exists such that each process can still get their maximum in sequence

• conceptually the processes could be run in this order

Deadlock: Example of Avoidance [117]

• assume that system has 12 tape drives

	maximum needs	current needs
P0	10	5
P1	4	2
P2	9	2
TOTAL	23	9

• sequence P1,P0,P2 is a safe sequence

• P2 requests one more tape drive:

	maximum needs	current needs
P0	10	5
P1	4	2
P2	9	3
TOTAL	23	10

• no safe sequence P1,... because P2 and P0 could deadlock

• for avoidance with multiple instances: use Banker's Algorithm

Avoidance with Single Instances [118]

SAFE:

UNSAFE (P2 requests R2): cycle

Deadlock: Detection [119]

• single instance resource types:

• periodically make a wait for graph

• (just remove resources from the allocation graph)

• check for cycles (n*n)

• multiple instance resource types:

• use a time-varying banker's algorithm

• how often should detection algorithm be run?

• how often is deadlock likely?

• how many processes will be affected?

Deadlock: Recovery [120]

• process termination

• abort all deadlocked processes

• abort one process at a time until cycle eliminated

• run detection algorithm each time

• which process to abort next?

• priority? CPU usage? how many resources?

• resource preemption

• take resources from some processes and give them to others

• select victim: cheapest cost?

• rollback: to safe state and restart

• starvation: deadlock could occur again, process restarts

• include number of rollbacks in cost

Deadlock: Combined Approach [121]

• resources belong to classes

• order the classes hierachically (each gets unique number)

• use appropriate technique within class

• Example of 4 classes:

• internal resources: for PCBs, etc.

• order the resources within the class

• central memory: for a user job

• preemption because a job can always be swapped out

• job resources: for tape drives, etc.

• avoidance because of job-control cards

• swapable space: for user jobs on backing store

• preallocation because maximum requirements known