Parallel Processing and
Concurrent Processes
Learning Objectives
After completing this chapter, you should be able to
describe:
The critical difference between processes and
processors, and their connection
The differences among common configurations of
multiprocessing systems
The significance of a critical region in process
synchronization
The basic concepts of process synchronization
software: test-and-set, WAIT and SIGNAL, and
semaphores
Understanding Operating Systems, Sixth Edition
Learning Objectives (cont'd.)
The need for process cooperation when several
processes work together
How several processors, executing a single job,
cooperate
The similarities and differences between processes
and threads
The significance of concurrent programming
languages and their applications
Understanding Operating Systems, Sixth Edition
What Is Parallel Processing?
Parallel processing
Multiprocessing
Two or more processors operate in unison
Two or more CPUs execute instructions
simultaneously
Processor Manager
Coordinates activity of each processor
Synchronizes interaction among CPUs
Understanding Operating Systems, Sixth Edition
What Is Parallel Processing? (cont'd.)
Parallel processing development
Enhances throughput
Increases computing power
Benefits
Increased reliability
More than one CPU
If one processor fails, others take over
Not simple to implement
Faster processing
Instructions processed in parallel two or more at a time
Understanding Operating Systems, Sixth Edition
What Is Parallel Processing? (cont'd.)
Faster instruction processing methods
CPU allocated to each program or job
CPU allocated to each working set or parts of it
Individual instructions subdivided
Each subdivision processed simultaneously
Concurrent programming
Two major challenges
Connecting processors into configurations
Orchestrating processor interaction
Synchronization is key
Understanding Operating Systems, Sixth Edition
What Is Parallel Processing? (cont'd.)
Understanding Operating Systems, Sixth Edition
Evolution of Multiprocessors
Developed for high-end midrange and mainframe
computers
Each additional CPU treated as additional resource
Today hardware costs reduced
Multiprocessor systems available on all systems
Multiprocessing occurs at three levels
Job level
Process level
Thread level
Each requires different synchronization frequency
Understanding Operating Systems, Sixth Edition
Evolution of Multiprocessors (cont'd.)
Understanding Operating Systems, Sixth Edition
Introduction to Multi-Core Processors
Multi-core processing
Several processors placed on single chip
Problems
Heat and current leakage (tunneling)
Solution
Single chip with two processor cores in same space
Allows two sets of simultaneous calculations
80 or more cores on single chip
Two cores each run more slowly than single core chip
Understanding Operating Systems, Sixth Edition
10
Multicore Era
Goal: to reach exascale computing
Motivation for Multicore Architectures
Limitation of performance improvement
Power wall (high power consumption and heat dissipation)
Memory wall (gap between processor speed and memory speed)
Frequency wall (high penalty for wrong branch prediction due to
deep pipelines with out-of-order execution)
For a uni-processor
Increase frequency by 20%
- Will increase power by more than 50%
- But increase performance by only 13%
Multicores with simple cores
decrease frequency, power consumption, and improve
performance
The Rise of Multicores
2001
, POWER4, IBM, 2 Cores
2005, Ultra SPARC T1
, Sun, 8 Cores
2008
, Core i7, Intel, 2/ 4/ 6 Cores
2012, Tesla GPGPU, Nvidia, 10 Cores
2012, Am2045, Ambric, a 336-core Massively
Parallel Processor Array (MPPA)
2014, Adapteva Epiphany, a many-core
processor architecture with up to 4096
processors on-chip
Top500.org
Started in 1993
Detects trends in high-performance computing (HPC)
Every 2 years sites operating the 500 most powerful
computer systems is resealed
How to rank the computer systems?
- Linpack benchmark (performance measure based
on floating point computing power, FLOPS)
Presented at the International Supercomputing
Conference
http://top500.org/
Typical Multiprocessing Configurations
Multiple processor configuration impacts
systems
Three types
Master/slave
Loosely coupled
Symmetric
Understanding Operating Systems, Sixth Edition
14
Master/Slave Configuration
Asymmetric multiprocessing system
Single-processor system
Additional slave processors
Each managed by primary master processor
Master processor responsibilities
Manages entire system
Maintains all processor status
Performs storage management activities
Schedules work for other processors
Executes all control programs
Understanding Operating Systems, Sixth Edition
15
Master/Slave Configuration (cont'd.)
Understanding Operating Systems, Sixth Edition
16
Master/Slave Configuration (cont'd.)
Advantages
Simplicity
Disadvantages
Reliability
No higher than single processor system
Potentially poor resources usage
Increases number of interrupts
Understanding Operating Systems, Sixth Edition
17
Loosely Coupled Configuration
Several complete computer systems
Each with own resources
Maintains commands and I/O management tables
Independent single-processing difference
Each processor
Communicates and cooperates with others
Has global tables
Several requirements and policies for job scheduling
Single processor failure
Others continue work independently
Difficult to detect
Understanding Operating Systems, Sixth Edition
18
Loosely Coupled Configuration (cont'd.)
Understanding Operating Systems, Sixth Edition
19
Symmetric Configuration
Decentralized processor scheduling
Each processor is same type
Advantages (over loosely coupled configuration)
More reliable
Uses resources effectively
Can balance loads well
Can degrade gracefully in failure situation
Most difficult to implement
Requires well synchronized processes
Avoids races and deadlocks
Understanding Operating Systems, Sixth Edition
20
Symmetric Configuration (cont'd.)
Understanding Operating Systems, Sixth Edition
21
Symmetric Configuration (cont'd.)
Decentralized process scheduling
Single operating system copy
Global table listing
Interrupt processing
Update corresponding process list
Run another process
More conflicts
Several processors access same resource at same
time Race condition
Process synchronization
Algorithms resolving conflicts between processors
Understanding Operating Systems, Sixth Edition
22
Process Synchronization Software
Successful process synchronization
Lock up used resource
Protect from other processes until released
Only when resource is released
Waiting process is allowed to use resource
Mistakes in synchronization can result in:
Starvation
Leave job waiting indefinitely
Deadlock
If key resource is being used
Understanding Operating Systems, Sixth Edition
23
Process Synchronization Software
(cont'd.)
Critical region
Part of a program
Critical region must complete execution
Other processes must wait before accessing critical
region resources
Processes within critical region
Cannot be interleaved
Threatens integrity of operation
Understanding Operating Systems, Sixth Edition
24
Process Synchronization Software
(cont'd.)
Synchronization
Implemented as lock-and-key arrangement:
Process determines key availability
Process obtains key
Puts key in lock
Makes it unavailable to other processes
Types of locking mechanisms
Test-and-set
WAIT and SIGNAL
Semaphores
Understanding Operating Systems, Sixth Edition
25
Test-and-Set
Indivisible machine instruction
Executed in single machine cycle
If key available: set to unavailable
Actual key
Single bit in storage location: zero (free) or one (busy)
Before process enters critical region
Tests condition code using TS instruction
No other process in region
Process proceeds
Condition code changed from zero to one
P1 exits: code reset to zero, allowing others to enter
Understanding Operating Systems, Sixth Edition
26
Test-and-Set (cont'd.)
Advantages
Simple procedure to implement
Works well for small number of processes
Drawbacks
Starvation
Many processes waiting to enter a critical region
Processes gain access in arbitrary fashion
Busy waiting
Waiting processes remain in unproductive, resourceconsuming wait loops
Understanding Operating Systems, Sixth Edition
27
WAIT and SIGNAL
Modification of test-and-set
Designed to remove busy waiting
Two new mutually exclusive operations
WAIT and SIGNAL
Part of process schedulers operations
WAIT
Activated when process encounters busy condition
code
SIGNAL
Activated when process exits critical region and
condition code set to free
Understanding Operating Systems, Sixth Edition
28
Semaphores
Nonnegative integer variable
Flag
Signals if and when resource is free
Resource can be used by a process
Two operations of semaphore
P (proberen means to test)
V (verhogen means to increment)
Understanding Operating Systems, Sixth Edition
29
Semaphores (cont'd.)
Understanding Operating Systems, Sixth Edition
30
Semaphores (cont'd.)
Let s be a semaphore variable
V(s): s: = s + 1
Fetch, increment, store sequence
P(s): If s > 0, then s: = s 1
Test, fetch, decrement, store sequence
s = 0 implies busy critical region
Process calling on P operation must wait until s > 0
Waiting job of choice processed next
Depends on process scheduler algorithm
Understanding Operating Systems, Sixth Edition
31
Semaphores (cont'd.)
Understanding Operating Systems, Sixth Edition
32
Semaphores (cont'd.)
P and V operations on semaphore s
Enforce mutual exclusion concept
Semaphore called mutex (MUTual EXclusion)
P(mutex): if mutex > 0 then mutex: = mutex 1
V(mutex): mutex: = mutex + 1
Critical region
Ensures parallel processes modify shared data only
while in critical region
Parallel computations
Mutual exclusion explicitly stated and maintained
Understanding Operating Systems, Sixth Edition
33
Process Cooperation
Several processes work together to complete
common task
Each case requires
Mutual exclusion and synchronization
Absence of mutual exclusion and synchronization
Results in problems
Examples
Producers and consumers problem
Readers and writers problem
Each case implemented using semaphores
Understanding Operating Systems, Sixth Edition
34
Producers and Consumers
One process produces data
Another process later consumes data
Example: CPU and line printer buffer
Delay producer: buffer full
Delay consumer: buffer empty
Implemented by two semaphores
Number of full positions
Number of empty positions
Mutex
Third semaphore: ensures mutual exclusion
Understanding Operating Systems, Sixth Edition
35
Producers and Consumers (cont'd.)
Understanding Operating Systems, Sixth Edition
36
Producers and Consumers (cont'd.)
Understanding Operating Systems, Sixth Edition
37
Producers and Consumers (cont'd.)
Understanding Operating Systems, Sixth Edition
38
Producers and Consumers (cont'd.)
Producers and Consumers Algorithm
empty: = n
full: = 0
mutex: = 1
COBEGIN
repeat until no more data PRODUCER
repeat until buffer is empty
CONSUMER
COEND
Understanding Operating Systems, Sixth Edition
39
Readers and Writers
Two process types need to access shared resource
Example: file or database
Example: airline reservation system
Implemented using two semaphores
Ensures mutual exclusion between readers and writers
Resource given to all readers
Provided no writers are processing (W2 = 0)
Resource given to a writer
Provided no readers are reading (R2 = 0) and no
writers writing (W2 = 0)
Understanding Operating Systems, Sixth Edition
40
Concurrent Programming
Concurrent processing system
One job uses several processors
Executes sets of instructions in parallel
Requires programming language and computer
system support
Understanding Operating Systems, Sixth Edition
41
Applications of Concurrent
Programming
A = 3 * B * C + 4 / (D + E) **(F G)
Understanding Operating Systems, Sixth Edition
42
Applications of Concurrent
Programming (cont'd.)
A = 3 * B * C + 4 / (D + E) ** (F G)
Understanding Operating Systems, Sixth Edition
43
Applications of Concurrent
Programming (cont'd.)
Two types of parallelism
Explicit parallelism
Requires programmer intervention
Explicitly state parallel executable instructions
Disadvantages
Time-consuming coding
Missed opportunities for parallel processing
Errors
Parallel processing mistakenly indicated
Programs difficult to modify
Understanding Operating Systems, Sixth Edition
44
Applications of Concurrent
Programming (cont'd.)
Implicit parallelism
Compiler automatically detects parallel instructions
Advantages
Solves explicit parallelism problems
Complexity dramatically reduced
Working with array operations within loops
Performing matrix multiplication
Conducting parallel searches in databases
Sorting or merging file
Understanding Operating Systems, Sixth Edition
45
Threads and Concurrent Programming
Threads
Small unit within process
Scheduled and executed
Minimizes overhead
Swapping process between main memory and
secondary storage
Each active process thread
Processor registers, program counter, stack and
status
Shares data area and resources allocated to its
process
Understanding Operating Systems, Sixth Edition
46
Thread States
Understanding Operating Systems, Sixth Edition
47
Thread States (cont'd.)
Operating system support
Creating new threads
Setting up thread
Ready to execute
Delaying or putting threads to sleep
Specified amount of time
Blocking or suspending threads
Those waiting for I/O completion
Setting threads to WAIT state
Until specific event occurs
Understanding Operating Systems, Sixth Edition
48
Thread States (cont'd.)
Operating system support (cont'd.)
Scheduling thread execution
Synchronizing thread execution
Using semaphores, events, or conditional variables
Terminating thread
Releasing its resources
Understanding Operating Systems, Sixth Edition
49
Thread Control Block
Information about current status and characteristics
of thread
Understanding Operating Systems, Sixth Edition
50
Summary
Multiprocessing
Single-processor systems
Interacting processes obtain control of CPU at different
times
Systems with two or more CPUs
Control synchronized by processor manager
Processor communication and cooperation
System configuration
Master/slave, loosely coupled, symmetric
Understanding Operating Systems, Sixth Edition
51
Summary (cont'd.)
Multiprocessing system success
Synchronization of resources
Mutual exclusion
Prevents deadlock
Maintained with test-and-set, WAIT and SIGNAL, and
semaphores (P, V, and mutex)
Synchronize processes using hardware and
software mechanisms
Understanding Operating Systems, Sixth Edition
52
Summary (cont'd.)
Avoid typical problems of synchronization
Missed waiting customers
Synchronization of producers and consumers
Mutual exclusion of readers and writers
Concurrent processing innovations
Threads and multi-core processors
Requires modifications to operating systems
Understanding Operating Systems, Sixth Edition
53
