1 Concurrency: Deadlock and Starvation - PowerPoint Presentation

Slide1

1

Concurrency: Deadlock and Starvation

Chapter 6Slide2

2

Outline

Principles of Deadlock

Introduction and conditions for deadlock

Deadlock prevention

Deadlock Avoidance

Deadlock detection

An Integrated deadlock strategy

Concurrency Mechanisms in UNIX, Linux, Solaris and WindowsSlide3

3

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

Some OS (ex: Unix SVR4) ignore the problem and pretend that deadlocks never occur...Slide4

4

Potential Deadlock

I need quad A and B

I need quad B and C

I need quad C and B

I need quad D and ASlide5

5

Actual Deadlock

HALT

until B is free

HALT

until C is free

HALT

until D is free

HALT

until A is freeSlide6

6

Two Processes P and Q

Let’s look at this with two processes P and Q.

Each needing exclusive access to a resource A and B for a period of time.

Deadlock could happen.

Whether or not deadlock occurs depends on both the dynamics of the execution and on the details of the application.Slide7

7

Example where deadlock can occurSlide8

8

Alternative logic

Suppose that P does not need both resources at the same time so that the two processes have this formSlide9

9

Example where deadlock cannot occurSlide10

10

The Conditions for Deadlock

These 3 conditions of policy must be present for a deadlock to be possible:

1

:

Mutual exclusion

only one process may use a resource at a time

2:

Hold-and-wait

a process may hold allocated resources while awaiting assignment of others

3:

No preemption

no resource can be forcibly removed from a process holding itSlide11

11

The Conditions for Deadlock

We also need the occurrence of a particular sequence of events that result in:

4:

Circular wait

a closed chain of processes exists, such that each process holds at least one resource needed by the next process in the chainSlide12

12

The Conditions for Deadlock

Deadlock occurs if and only if the circular wait condition is unresolvable

The circular wait condition is unresolvable when the first 3 policy conditions hold

Thus the 4 conditions taken together constitute necessary and sufficient conditions for deadlockSlide13

13

Resource Categories

Two general categories of resources:

Reusable

can be safely used by only one process at a time and

is not depleted

by that use.

Example?

Consumable

one that can be created (

produced

) and destroyed (

consumed

).Example?Slide14

14

Reusable Resources

Such as:

Processors, I/O channels, main and secondary memory, devices, and data structures such as files, databases, and semaphores

Deadlock

may

occur if each process holds one resource and requests the otherSlide15

15

Example of Reuse Deadlock

Consider two processes that compete for exclusive access to a disk file D and a tape drive T.

Deadlock occurs if each process holds one resource and requests the other.Slide16

16

Reusable Resources Example 1: Device RequestSlide17

17

Reusable Resources Example 2: Memory Request

Space is available for allocation of 200Kbytes, and the following sequence of events occur

Deadlock occurs if both processes progress to their second request (assuming no-preemption)

P1

. . .

. . .

Request 80 Kbytes;

Request 60 Kbytes;

P2

. . .

. . .

Request 70 Kbytes;

Request 80 Kbytes;Slide18

18

Consumable Resources

Such as Interrupts, signals, messages, and information in I/O buffers

Deadlock may occur if a Receive message is blocking

May take a rare combination of events to cause deadlockSlide19

19

Example of Deadlock

Consider a pair of processes, in which each process attempts to receive a message from the other process and then send a message to the other processSlide20

20

Resource Allocation Graphs

Directed graph that depicts a state of the system of resources and processesSlide21

21

Methods for handling deadlocks

Deadlock prevention

disallow 1 of the 4 necessary conditions of deadlock occurrence

Deadlock avoidance

do not grant a resource request if this allocation

might

lead to deadlock

Deadlock detection

always grant resource request when possible. But periodically check for the presence of deadlock and then recover from itSlide22

22

Deadlock Prevention

The OS is design in such a way as to exclude a priori the possibility of deadlock

Two main methods:

Indirect - disallow 1of the 3 policy conditions or prevent all three of the necessary conditions occurring at once

Direct - prevent the occurrence of circular waitSlide23

23

Indirect methods of deadlock prevention

Mutual Exclusion

cannot be disallowed

ex: only 1 process at a time can write to a file

Hold-and-Wait

can be disallowed by requiring that a process request all its required resources at one time

block the process until all requests can be granted simultaneously

Issues of Hold-&-Wait:

process may be held up for a long time waiting for all its requests

resources allocated to a process may remain unused for a long time. These resources could be used by other processes

an application would need to be aware of all the resources that will be neededSlide24

24

Indirect methods of deadlock prevention

No preemption

Process must release resource and request again.

But whenever a process must release a resource who’s usage is in progress, the state of this resource must be saved for later resumption.

Hence: practical only when the state of a resource can be easily saved and restored later, such as the processor.Slide25

25

Direct methods of deadlock prevention

A protocol to prevent circular wait:

define a strictly increasing linear ordering O() for resource types. Ex:

R1: tape drives: O(R1) = 2

R2: disk drives: O(R2) = 4

R3: printers: O(R3) = 7

A process initially request a number of instances of a resource type, say Ri. A single request must be issued to obtain several instances.

After that, the process can request instances for resource type Rj if and only if O(Rj) > O(Ri)Slide26

26

Prevention of circular wait

Circular wait cannot hold under this protocol. Proof:

Processes {P0, P1..Pn} are involved in circular wait iff Pi is waiting for Ri which is held by Pi+1 and Pn is waiting for Rn held which is held by P0 (circular waiting)Slide27

27

Prevention of circular wait

under this protocol, this means:

O(R0) < O(R1) < .. < O(

Rn

) < O(R0) impossible!

This protocol prevents deadlock but will often deny resources unnecessarily (inefficient) because of the ordering imposed on the requestsSlide28

28

Deadlock Prevention: Summary

We disallow 1of the 3 policy conditions or use a protocol that prevents circular wait

This leads to

inefficient

use of resources and inefficient execution of processesSlide29

29

Deadlock Avoidance

We allow the 3 policy conditions but dynamically make a decision

whether the current resource allocation request will, if granted,

potentially lead to a deadlock

. If so, t

wo approaches:

Process initial denial: do not start a process if it’s demand might lead to deadlock

Resource allocation denial:

do not grant an incremental resource request if this allocation might lead to deadlock

Assumptions:

Maximum requirements

of each resource must be stated in advance

Requires

knowledge of future process requests

Allows more concurrency than preventionSlide30

30

Resource types

Resources in a system are partitioned in

resources types

Each resource type in a system exists with a certain amount. Let R(

i

) be the total amount of resource type

i

present in the system. Ex:

R(main memory) = 128 MB

R(disk drives) = 8

R(printers) = 5

The partition is system specific (ex: printers may be further partitioned...)Slide31

31

Process initiation denial

A process is only started if the maximum claim of all current processes plus those of the new process can be met.

Let C(

k,i

) be the amount of resource type

i

claimed

by process k.

To be admitted in the system, process k must show C(

k,i

) for all resource types

i

C(

k,i

) is the maximum value of resource type

i permitted for process k.Let U(

i) be the total amount of resource type i unclaimed

(available) in the system:U(i) = R(i

) -

S

_

k

C(

k,i

)Slide32

32

Process initiation denial

A new process n is admitted in the system only if C(

n,i

) <= U(

i

) for all resource type I

This policy ensures that deadlock is always avoided since a process is admitted only if all its requests can always be satisfied (no matter what will be their order)

A

sub optimal strategy

since it assumes the worst: that all processes will make their

maximum claims together at the same time Slide33

33

Resource allocation denial: the banker’s algorithm

Processes are like customers wanting to borrow money (resources) from a bank ...

A banker should not allocate cash when it cannot satisfy the needs of all its customers

Consider a system with a fixed number of resources

State

of the system is the current allocation of resources to process:

The values of R(

i

), C(

j,i

) for all resource type

i

and process j and the values of other vectors and matrices.

Safe state

is where there is

at least one sequence

that does not result in deadlock

Unsafe state is a state that is not safeSlide34

34

Determination of Safe State

A system consisting of four processes and three resources.

Allocations are made to processors

Is this a safe state?

Can any of the four processes be run to completion with the resources available?

Amount of Existing

Resources

Resources available after allocationSlide35

35

After P2

runs to completion

Can any of the remaining processes be completed?

Note P2 is completedSlide36

36

After P1 completesSlide37

37

P3 Completes

Finally, we can complete P4. At this point,

all

of the processes have been run to completion.

Thus, the state defined originally is a safe state.Slide38

38

Determination of an Unsafe State

This time Suppose that P1 makes the request for one additional unit each of R1 and R3.

Is this safe?Slide39

39

Deadlock Avoidance

When a process makes a request for a set of resources,

assume that the request is granted,

Update the system state accordingly,

Then determine if the result is a safe state.

If so, grant the request and,

if not, block the process until it is safe to grant the request.Slide40

40

The banker’s algorithm

We also need the amount

allocated

A(j,i) of resource type i to process j for all (j,i)

The total amount

available

of resource i is given by: V(i) = R(i) -

S

_k

A(k,i)

We also use the

need

N(j,i) of resource type i required by process j to complete its task: N(j,i) = C(j,i) - A(j,i)

To decide if a resource request made by a process should be granted, the banker’s algorithm test if granting the request will lead to a

safe state

:

If the resulting state is safe then grant request Else do not grant the requestSlide41

41

The banker’s algorithm

A state is safe iff there exists a sequence {P1..Pn} where each Pi is allocated all of its needed resources to be run to completion

i.e.: we can always run all the processes to completion from a safe state

The

safety algorithm

is the part that determines if a state is safe

Initialization:

all processes are said to be “unfinished”

set the work vector to the amount resources available: W(i) = V(i) for all i; Slide42

42

The banker’s algorithm

REPEAT: Find a unfinished process j such that N(j,i) <= W(i) for all i.

If no such j exists, goto EXIT

Else: “finish” this process and recover its resources: W(i) = W(i) + A(j,i) for all i. Then goto REPEAT

EXIT: If all processes have “finished” then this state is safe. Else it is unsafe.Slide43

43

The banker’s algorithm

Let Q(j,i) be the amount of resource type i requested by process j.

To determine if this request should be granted we use the

banker’s algorithm

:

If Q(j,i) <= N(j,i) for all i then continue. Else raise error condition (claim exceeded).

If Q(j,i) <= V(i) for all i then continue. Else wait (resource not yet available)

Pretend that the request is granted and determine the new resource-allocation state: Slide44

44

The banker’s algorithm

V(i) = V(i) - Q(j,i) for all i

A(j,i) = A(j,i) + Q(j,i) for all i

N(j,i) = N(j,i) - Q(j,i) for all i

If the resulting state is safe then allocate resource to process j. Else process j must wait for request Q(j,i) and restore old state. Slide45

45

Banker’s algorithm: comments

A safe state cannot be deadlocked. But an unsafe state is not necessarily deadlocked.

Ex: P1 from the previous (unsafe) state could release temporarily a unit of R1 and R3 (returning to a safe state)

some processes may need to wait unnecessarily

sub optimal use of resourcesSlide46

46

Deadlock Avoidance Restrictions

Maximum

resource requirement must be stated in advance

Processes under consideration must be

independent

and with no synchronization requirements

Example of using shared memory (SM) among multiple processes: SM is counted multiple times

There must be a

fixed

number of resources to allocate

No process may exit

while holding resourcesSlide47

47

Deadlock Detection

Resource access are

granted

to processes

whenever possible

. The OS needs:

an algorithm to check if deadlock is present

an algorithm to recover from deadlock

The deadlock check can be performed at every resource request

Such frequent checks consume CPU timeSlide48

48

A deadlock detection algorithm

Makes use of previous resource-allocation matrices and vectors

Marks each process not deadlocked. Initially all processes are unmarked. Then perform:

Mark each process j for which: A(j,i) = 0 for all resource type i. (since these are not deadlocked)

Initialize work vector: W(i) = V(i) for all i

REPEAT: Find a unmarked process j such that Q(j,i) <= W(i) for all i. Stop if such j does not exists.

If such j exists: mark process j and set W(i) = W(i) + A(j,i) for all i. Goto REPEAT

At the end: each unmarked process is deadlockedSlide49

49

Deadlock detection: comments

Process j is not deadlocked when Q(

j,i

) <= W(

i

) for all

i

.

Then we are

optimistic

and

assume

that process j will require no more resources to complete its task.

It will thus soon return all of its allocated resources. Thus: W(

i

) = W(i

) + A(j,i) for all i.

If this assumption is incorrect, a deadlock may occur later.This deadlock will be detected the next time the deadlock detection algorithm is invoked.Slide50

50

Deadlock detection: example

What is R?

Mark P4 since it has no allocated resources

Set W = (0,0,0,0,1)

P3’s request <= W. So mark P3 and set W = W + (0,0,0,1,0) = (0,0,0,1,1)

Algorithm terminates. P1 and P2 are deadlocked. Why?

R1 R2 R3 R4 R5

P1

P2

P3

P4

Request Allocated Available

R1 R2 R3 R4 R5

R1 R2 R3 R4 R5

0 1 0 0 1

0 0 1 0 1

0 0 0 0 1

1 0 1 0 1

1 0 1 1 0

1 1 0 0 0

0 0 0 1 0

0 0 0 0 0

0 0 0 0 1Slide51

51

Deadlock Recovery

Needed when deadlock is detected. The following approaches are possible:

Abort all deadlocked processes (one of the most common solutions adopted in OS!!)

Rollback each deadlocked process to some previously defined checkpoint and restart them (original deadlock may reoccur)

Successively abort deadlocked processes until deadlock no longer exists (each time we need to invoke the deadlock detection algorithm)Slide52

52

Deadlock Recovery (cont.)

Successively preempt some resources from processes and give them to other processes until deadlock no longer exists

a process that has a resource preempted must be rolled back prior to its acquisition

For the 2 last approaches: a victim process needs to be selected according to:

least amount of CPU time consumed so far

least total resources allocated so far

least amount of “work” produced so far

Lowest priority

…Slide53

53

An integrated deadlock strategy

We can combine the previous approaches into the following way:

Group resources into a number of different classes and order them. Ex:

Swappable space (secondary memory)

Process resources (I/O devices, files...)

Main memory...

Use prevention of circular wait to prevent deadlock between resource classes

Use the most appropriate approach for each class for deadlocks within each class Slide54

54

Summary - Deadlock

Detection,

Prevention, and AvoidanceSlide55

55

Roadmap

Principals of Deadlock

Deadlock prevention

Deadlock Avoidance

Deadlock detection

An Integrated deadlock strategy

Concurrency Mechanisms in UNIX and LinuxSlide56

56

UNIX / Linux Concurrency Mechanisms

UNIX/ Linux provides a variety of mechanisms for

interprocess

communications and synchronization including:

Pipes (FIFOs)

Messages

Shared memory

Semaphores

SignalsSlide57

57

Pipes

A circular buffer allowing two processes to communicate on the producer-consumer model

first-in-first-out queue, written by one process and read by another.

Two types:

Named:

UnnamedSlide58

58

Messages

A block of bytes with an accompanying type.

UNIX provides

msgsnd

and

msgrcv

system calls for processes to engage in message passing.

Associated with each process is a message queue, which functions like a mailbox.Slide59

59

Shared Memory

A common block of virtual memory shared by multiple processes.

Permission is read-only or read-write for a process,

determined on a per-process basis.

Mutual exclusion constraints are not part of the shared-memory facility but must be provided by the processes using the shared memory.Slide60

60

Semaphores

SVR4 uses a generalization of the

semWait

and

semSignal

primitives defined in Chapter 5;

Associated with the semaphore are queues of processes blocked on that semaphore.Slide61

61

Signals

A software mechanism that informs a process of the occurrence of asynchronous events.

Similar to a hardware interrupt, without priorities

A signal is delivered by updating a field in the process table for the process to which the signal is being sent.Slide62

62

Linux Kernel Concurrency Mechanism

Includes all the mechanisms found in UNIX plus

Atomic operations

Spinlocks

Semaphores (slightly different from SVR4)

BarriersSlide63

63

Atomic Operations

Atomic operations execute without interruption and without interference

Two types:

Integer operations – operating on an integer variable

Bitmap operations – operating on one bit in a bitmap Slide64

64

Linux Atomic OperationsSlide65

65

Linux Atomic OperationsSlide66

66

Spinlock

Only one thread at a time can acquire a spinlock.

Any other thread will keep trying (spinning) until it can acquire the lock.

A spinlock is an integer

If 0, the thread sets the value to 1 and enters its critical section.

If the value is nonzero, the thread continually checks the value until it is zero.

Disadvantage?

Why

using spinlocks?Slide67

67

Semaphores

User level:

Linux provides a semaphore interface corresponding to that in UNIX

SVR4

Internally:

implemented as functions within the kernel and are more efficient than

user-

visable

semaphores

Three types of kernel semaphores:

binary semaphores

counting semaphores

reader-writer semaphoresSlide68

68

Linux

Semaphores