1

CMP 252 – PARALLELISM FUNDAMENTALS (2 UNITS)

Course Outline
Multiple simultaneous computations: Goals of parallelism (e.g., throughput) versus concurrency
(e.g., controlling access to shared resources); Parallelism, communication, and coordination
Programming constructs for coordinating multiple simultaneous computations.
Need for synchronization; Programming errors not found in sequential programming Data races
(simultaneous read/write or write/write of shared state); Higher-level races (inter-leavings
violating program intention, undesired non-determinism); Lack of live-ness/progress (deadlock,
starvation).
Need for communication and coordination/synchronization: Independence and partitioning;
Shared Memory: Consistency and its role in programming language guarantees for data-race-free
programs
Multicore processors: Shared vs. distributed memory

Text:
Introduction to Parallel Computing
Author: Blaise Barney, Lawrence Livermore National Laboratory
 2

Introduction to Parallel Computing

Author: Blaise Barney, Lawrence Livermore National Laboratory

Table of Contents
1. Abstract
2. Overview
1. What is Parallel Computing?
2. Why Use Parallel Computing?
3. Who is Using Parallel Computing?
3. Concepts and Terminology
1. von Neumann Computer Architecture
2. Flynn's Classical Taxonomy
3. Some General Parallel Terminology
4. Limits and Costs of Parallel Programming
4. Parallel Computer Memory Architectures
1. Shared Memory
2. Distributed Memory
3. Hybrid Distributed-Shared Memory
5. Parallel Programming Models
1. Overview
2. Shared Memory Model
3. Threads Model
4. Distributed Memory / Message Passing Model
5. Data Parallel Model
6. Hybrid Model
7. SPMD and MPMP
6. Designing Parallel Programs
1. Automatic vs. Manual Parallelization
2. Understand the Problem and the Program
3. Partitioning
4. Communications
5. Synchronization
6. Data Dependencies
7. Load Balancing
8. Granularity
9. I/O
10. Debugging
11. Performance Analysis and Tuning
7. Parallel Examples
1. Array Processing
2. PI Calculation
3. Simple Heat Equation
4. 1-D Wave Equation
8. References and More Information
 3

Abstract
This tutorial is intended to provide only a very quick overview of the extensive and broad topic of
Parallel Computing. As such, it covers just the very basics of parallel computing, and is intended for
someone who is just becoming acquainted with the subject and who is planning to attend one or more
other tutorials. It is not intended to cover Parallel Programming in depth, as this would require
significantly more time. The tutorial begins with a discussion on parallel computing - what it is and how
it's used, followed by a discussion on concepts and terminology associated with parallel computing.
The topics of parallel memory architectures and programming models are then explored. These topics
are followed by a series of practical discussions on a number of the complex issues related to
designing and running parallel programs. The tutorial concludes with several examples of how to
parallelize simple serial programs.

What is Parallel Computing?

Serial Computing:

 Traditionally, software has been written for serial computation:

o A problem is broken into a discrete series of instructions
o Instructions are executed sequentially one after another
o Executed on a single processor
o Only one instruction may execute at any moment in time
 4

For example:

Parallel Computing:

 In the simplest sense, parallel computing is the simultaneous use of multiple computing
resources to solve a computational problem:
o A problem is broken into discrete parts that can be solved concurrently
o Each part is further broken down to a series of instructions
o Instructions from each part execute simultaneously on different processors
o An overall control/coordination mechanism is employed
 5

For example:

 The computational problem should be able to:

o Be broken apart into discrete pieces of work that can be solved simultaneously;
o Execute multiple program instructions at any moment in time;
o Be solved in less time with multiple computing resources than with a single computing
resource.
 The computing resources are typically:
o A single computer with multiple processors/cores
o An arbitrary number of such computers connected by a network

Parallel Computers:

 Virtually all stand-alone computers today are parallel from a hardware perspective:
o Multiple functional units (L1 cache, L2 cache, branch, prefetch, decode, floating-point,
graphics processing (GPU), integer, etc.)
o Multiple execution units/cores
o Multiple hardware threads
 6

IBM BG/Q Compute Chip with 18 cores (PU) and 16 L2 Cache units (L2)

 Networks connect multiple stand-alone computers (nodes) to make larger parallel computer
clusters.

 For example, the schematic below shows a typical LLNL parallel computer cluster:
o Each compute node is a multi-processor parallel computer in itself
o Multiple compute nodes are networked together with an Infiniband network
o Special purpose nodes, also multi-processor, are used for other purposes
 7

 The majority of the world's large parallel computers (supercomputers) are clusters of hardware
produced by a handful of (mostly) well known vendors.
 8

Source: Top500.org

Why Use Parallel Computing?

The Real World is Massively Parallel:
 In the natural world, many complex, interrelated events are happening at the same time, yet within a
temporal sequence.
 Compared to serial computing, parallel computing is much better suited for modeling, simulating and
understanding complex, real world phenomena.
 For example, imagine modeling these serially:
 9

Main Reasons:

 SAVE TIME AND/OR MONEY:

o In theory, throwing more resources at a task will shorten its time to completion, with potential
cost savings.
o Parallel computers can be built from cheap, commodity components.
 10

 SOLVE LARGER / MORE COMPLEX PROBLEMS:

o Many problems are so large and/or complex that it is impractical or impossible to solve them
on a single computer, especially given limited computer memory.
o Example: "Grand Challenge Problems" (en.wikipedia.org/wiki/Grand_Challenge) requiring
PetaFLOPS and PetaBytes of computing resources.
o Example: Web search engines/databases processing millions of transactions every second

 PROVIDE CONCURRENCY:
o A single computing resource can only do one thing at a time. Multiple computing resources
can do many things simultaneously.
o Example: Collaborative Networks provide a global venue where people from around the
world can meet and conduct work "virtually".
 11

 TAKE ADVANTAGE OF NON-LOCAL RESOURCES:

o Using computing resources on a wide area network, or even the Internet when local
computing resources are scarce or insufficient.
o Example: SETI@home (setiathome.berkeley.edu) over 1.5 million users in nearly every
country in the world. Source: www.boincsynergy.com/stats/ (June, 2015).
o Example: Folding@home (folding.stanford.edu) uses over 160,000 computers globally (June,
2015)

 MAKE BETTER USE OF UNDERLYING PARALLEL HARDWARE:

o Modern computers, even laptops, are parallel in architecture with multiple processors/cores.
o Parallel software is specifically intended for parallel hardware with multiple cores, threads,
etc.
o In most cases, serial programs run on modern computers "waste" potential computing power.

Intel Xeon processor with 6 cores and 6 L3 cache units

The Future:

 During the past 20+ years, the trends indicated by ever faster networks, distributed systems, and
multi-processor computer architectures (even at the desktop level) clearly show that parallelism is the
future of computing.
 In this same time period, there has been a greater than 500,000x increase in supercomputer
performance, with no end currently in sight.
 12

 The race is already on for Exascale Computing!

18
o Exaflop = 10 calculations per second

Source: Top500.org
 13

Who is Using Parallel Computing?

Science and Engineering:

 Historically, parallel computing has been considered to be "the high end of computing", and has been used
to model difficult problems in many areas of science and engineering:

o Atmosphere, Earth, Environment o Mechanical Engineering - from

o Physics - applied, nuclear, prosthetics to spacecraft
particle, condensed matter, high o Electrical Engineering, Circuit Design,
pressure, fusion, photonics Microelectronics
o Bioscience, Biotechnology, o Computer Science, Mathematics
Genetics o Defense, Weapons
o Chemistry, Molecular Sciences
o Geology, Seismology

Industrial and Commercial:

 Today, commercial applications provide an equal or greater driving force in the development of faster
computers. These applications require the processing of large amounts of data in sophisticated ways. For
example:

o "Big Data", databases, data mining o Financial and economic modeling

o Oil exploration o Management of national and multi-national
o Web search engines, web based corporations
business services o Advanced graphics and virtual reality, particularly in
o Medical imaging and diagnosis the entertainment industry
o Pharmaceutical design o Networked video and multi-media technologies
o Collaborative work environments
 14

Global Applications:

 Parallel computing is now being used extensively around the world, in a wide variety of
applications.

Source: Top500.org
 15

von Neumann Architecture

 Named after the Hungarian mathematician/genius John von Neumann who first authored the
general requirements for an electronic computer in his 1945 papers.
 Also known as "stored-program computer" - both program instructions and data are kept in
electronic memory. Differs from earlier computers which were programmed through "hard
wiring".
 Since then, virtually all computers have followed this basic design:

 Comprised of four main

components:
o Memory
o Control Unit
o Arithmetic Logic
Unit
o Input/Output
 Read/write, random access
memory is used to store both
program instructions and
data
o Program instructions
are coded data which
tell the computer to
do something
o Data is simply
information to be
used by the program John von Neumann circa 1940s
 Control unit fetches (Source: LANL archives)
instructions/data from
memory, decodes the
instructions and then
sequentially coordinates
operations to accomplish the
programmed task.
 Arithmetic Unit performs
basic arithmetic operations
 Input/Output is the interface
to the human operator

 More info on his other remarkable accomplishments:

http://en.wikipedia.org/wiki/John_von_Neumann
 So what? Who cares?
o Well, parallel computers still follow this basic design, just multiplied in units. The basic,
fundamental architecture remains the same.
 16

Flynn's Classical Taxonomy

 There are different ways to classify parallel computers. Examples available HERE.
 One of the more widely used classifications, in use since 1966, is called Flynn's Taxonomy.
 Flynn's taxonomy distinguishes multi-processor computer architectures according to how they
can be classified along the two independent dimensions of Instruction Stream and Data
Stream. Each of these dimensions can have only one of two possible states: Single or Multiple.
 The matrix below defines the 4 possible classifications according to Flynn:

Single Instruction, Single Data (SISD):

 A serial (non-parallel) computer

 Single Instruction: Only one instruction stream is being acted on by the CPU during any one
clock cycle
 Single Data: Only one data stream is being used as input during any one clock cycle
 Deterministic execution
 This is the oldest type of computer
 Examples: older generation mainframes, minicomputers, workstations and single processor/core
PCs.
 17

UNIVAC1 IBM 360 CRAY1

PDP1
CDC 7600 Dell Laptop

Single Instruction, Multiple Data (SIMD):

 A type of parallel computer

 Single Instruction: All processing units execute the same instruction at any given clock cycle
 Multiple Data: Each processing unit can operate on a different data element
 Best suited for specialized problems characterized by a high degree of regularity, such as
graphics/image processing.
 Synchronous (lockstep) and deterministic execution
 Two varieties: Processor Arrays and Vector Pipelines
 Examples:
o Processor Arrays: Thinking Machines CM-2, MasPar MP-1 & MP-2, ILLIAC IV
o Vector Pipelines: IBM 9000, Cray X-MP, Y-MP & C90, Fujitsu VP, NEC SX-2, Hitachi
S820, ETA10
 18

 Most modern computers, particularly those with graphics processor units (GPUs) employ SIMD
instructions and execution units.

ILLIAC IV MasPar


Cray Y-MP Thinking Machines CM-2

Cray X-MP
Cell Processor (GPU)

Multiple Instruction, Single Data (MISD):

 A type of parallel computer

 Multiple Instruction: Each processing unit operates on the data independently via separate
instruction streams.
 Single Data: A single data stream is fed into multiple processing units.
 Few (if any) actual examples of this class of parallel computer have ever existed.
 Some conceivable uses might be:
o multiple frequency filters operating on a single signal stream
o multiple cryptography algorithms attempting to crack a single coded message.
 19

Multiple Instruction, Multiple Data (MIMD):

 A type of parallel computer

 Multiple Instruction: Every processor may be executing a different instruction stream
 Multiple Data: Every processor may be working with a different data stream
 Execution can be synchronous or asynchronous, deterministic or non-deterministic
 Currently, the most common type of parallel computer - most modern supercomputers fall into
this category.
 20

IBM POWER5 HP/Compaq Alphaserver

 Examples: most current supercomputers, networked parallel computer clusters and "grids",
multi-processor SMP computers, multi-core PCs.
 Note: many MIMD architectures also include SIMD execution sub-components


AMD Opteron Cray XT3 IBM BG/L

Some General Parallel Terminology

 Like everything else, parallel computing has its own "jargon". Some of the more commonly used
terms associated with parallel computing are listed below.
 Most of these will be discussed in more detail later.

Supercomputing / High Performance Computing (HPC)

Using the world's fastest and largest computers to solve large problems.
Node
A standalone "computer in a box". Usually comprised of multiple CPUs/processors/cores,
memory, network interfaces, etc. Nodes are networked together to comprise a supercomputer.
CPU / Socket / Processor / Core
This varies, depending upon who you talk to. In the past, a CPU (Central Processing Unit) was a
singular execution component for a computer. Then, multiple CPUs were incorporated into a
node. Then, individual CPUs were subdivided into multiple "cores", each being a unique
execution unit. CPUs with multiple cores are sometimes called "sockets" - vendor dependent.
The result is a node with multiple CPUs, each containing multiple cores. The nomenclature is
confused at times. Wonder why?
 21

Task
A logically discrete section of computational work. A task is typically a program or program-like
set of instructions that is executed by a processor. A parallel program consists of multiple tasks
running on multiple processors.
Pipelining
Breaking a task into steps performed by different processor units, with inputs streaming through,
much like an assembly line; a type of parallel computing.
Shared Memory
From a strictly hardware point of view, describes a computer architecture where all processors
have direct (usually bus based) access to common physical memory. In a programming sense, it
describes a model where parallel tasks all have the same "picture" of memory and can directly
address and access the same logical memory locations regardless of where the physical
memory actually exists.
Symmetric Multi-Processor (SMP)
Shared memory hardware architecture where multiple processors share a single address space
and have equal access to all resources.
Distributed Memory
In hardware, refers to network based memory access for physical memory that is not common.
As a programming model, tasks can only logically "see" local machine memory and must use
communications to access memory on other machines where other tasks are executing.
Communications
Parallel tasks typically need to exchange data. There are several ways this can be
accomplished, such as through a shared memory bus or over a network, however the actual
event of data exchange is commonly referred to as communications regardless of the method
employed.
Synchronization
The coordination of parallel tasks in real time, very often associated with communications. Often
implemented by establishing a synchronization point within an application where a task may not
proceed further until another task(s) reaches the same or logically equivalent point.
 22

Synchronization usually involves waiting by at least one task, and can therefore cause a parallel
application's wall clock execution time to increase.

Granularity
In parallel computing, granularity is a qualitative measure of the ratio of computation to
communication.

o Coarse: relatively large amounts of computational work are done between

communication events
o Fine: relatively small amounts of computational work are done between communication
events

Observed Speedup
Observed speedup of a code which has been parallelized, defined as:
wall-clock time of serial execution
-----------------------------------
wall-clock time of parallel execution

One of the simplest and most widely used indicators for a parallel program's performance.

Parallel Overhead
The amount of time required to coordinate parallel tasks, as opposed to doing useful work.
Parallel overhead can include factors such as:
o Task start-up time
o Synchronizations
o Data communications
o Software overhead imposed by parallel languages, libraries, operating system, etc.
o Task termination time

Massively Parallel
Refers to the hardware that comprises a given parallel system - having many processing
elements. The meaning of "many" keeps increasing, but currently, the largest parallel computers
are comprised of processing elements numbering in the hundreds of thousands to millions.

Embarrassingly Parallel
Solving many similar, but independent tasks simultaneously; little to no need for coordination
between the tasks.

Scalability
Refers to a parallel system's (hardware and/or software) ability to demonstrate a proportionate
increase in parallel speedup with the addition of more resources. Factors that contribute to
scalability include:

o Hardware - particularly memory-cpu bandwidths and network communication properties

o Application algorithm
o Parallel overhead related
o Characteristics of your specific application
 23

Limits and Costs of Parallel Programming

Amdahl's Law:
 Amdahl's Law states that potential program speedup is defined by the fraction of code (P) that can
be parallelized:

1
speedup = ------
1 - P

 If none of the code can be parallelized, P = 0 and the speedup = 1 (no speedup).
 If all of the code is parallelized, P = 1 and the speedup is infinite (in theory).
 If 50% of the code can be parallelized, maximum speedup = 2, meaning the code will run twice as
fast.
 Introducing the number of processors performing the parallel fraction of work, the relationship can
be modeled by:

1
speedup = ------------
P + S
---
N
 24

where P = parallel fraction, N = number of processors and S = serial fraction.

 It soon becomes obvious that there are limits to the scalability of parallelism. For example:

speedup
--------------------------------
N P = .50 P = .90 P = .99
----- ------- ------- -------
10 1.82 5.26 9.17
100 1.98 9.17 50.25
1,000 1.99 9.91 90.99
10,000 1.99 9.91 99.02
100,000 1.99 9.99 99.90

 However, certain problems demonstrate increased performance by increasing the problem size.
For example:

2D Grid Calculations 85 seconds 85%

Serial fraction 15 seconds 15%

We can increase the problem size by doubling the grid dimensions and halving the time step.
This results in four times the number of grid points and twice the number of time steps. The
timings then look like:
 25

2D Grid Calculations 680 seconds 97.84%

Serial fraction 15 seconds 2.16%

 Problems that increase the percentage of parallel time with their size are more scalable than
problems with a fixed percentage of parallel time.

Complexity:

 In general, parallel applications are much more complex than corresponding serial applications,
perhaps an order of magnitude. Not only do you have multiple instruction streams executing at
the same time, but you also have data flowing between them.
 The costs of complexity are measured in programmer time in virtually every aspect of the
software development cycle:
o Design
o Coding
o Debugging
o Tuning
o Maintenance
 Adhering to "good" software development practices is essential when working with parallel
applications - especially if somebody besides you will have to work with the software.

Portability:

 Thanks to standardization in several APIs, such as MPI, POSIX threads, and OpenMP,
portability issues with parallel programs are not as serious as in years past. However...
 All of the usual portability issues associated with serial programs apply to parallel programs. For
example, if you use vendor "enhancements" to Fortran, C or C++, portability will be a problem.
 Even though standards exist for several APIs, implementations will differ in a number of details,
sometimes to the point of requiring code modifications in order to effect portability.
 Operating systems can play a key role in code portability issues.
 Hardware architectures are characteristically highly variable and can affect portability.

Resource Requirements:

 The primary intent of parallel programming is to decrease execution wall clock time, however in
order to accomplish this, more CPU time is required. For example, a parallel code that runs in 1
hour on 8 processors actually uses 8 hours of CPU time.
 The amount of memory required can be greater for parallel codes than serial codes, due to the
need to replicate data and for overheads associated with parallel support libraries and
subsystems.
 For short running parallel programs, there can actually be a decrease in performance compared
to a similar serial implementation. The overhead costs associated with setting up the parallel
environment, task creation, communications and task termination can comprise a significant
portion of the total execution time for short runs.
 26

Scalability:

 Two types of scaling based on time to solution:

o Strong scaling: The total problem size stays fixed as more processors are added.
o Weak scaling: The problem size per processor stays fixed as more processors are
added.
 The ability of a parallel program's performance to scale is a result of a number of interrelated
factors. Simply adding more processors is rarely the answer.
 The algorithm may have inherent limits to scalability. At some point, adding more resources
causes performance to decrease. Many parallel solutions demonstrate this characteristic at
some point.
 Hardware factors play a significant role in scalability. Examples:
o Memory-cpu bus bandwidth on an SMP machine
o Communications network bandwidth
o Amount of memory available on any given machine or set of machines
o Processor clock speed
 Parallel support libraries and subsystems software can limit scalability independent of your
application.

Parallel Computer Memory Architectures

Shared Memory
General Characteristics:
 Shared memory parallel computers vary widely, but generally have in common the ability for all
processors to access all memory as global address space.
 Multiple processors can operate independently but share the same memory resources.
 Changes in a memory location effected by one processor are visible to all other processors.
 Historically, shared memory machines have been classified as UMA and NUMA, based upon
memory access times.

Uniform Memory Access (UMA):

 Most commonly represented today by Symmetric Multiprocessor (SMP) machines

 Identical processors
 Equal access and access times to memory
 Sometimes called CC-UMA - Cache Coherent UMA. Cache coherent means if one processor
updates a location in shared memory, all the other processors know about the update. Cache
coherency is accomplished at the hardware level.

Shared Memory (UMA)

 27

Non-Uniform Memory Access (NUMA):

 Often made by physically linking two or more SMPs

 One SMP can directly access memory of another SMP
 Not all processors have equal access time to all memories
 Memory access across link is slower
 If cache coherency is maintained, then may also be called CC-NUMA - Cache Coherent NUMA

Shared Memory (NUMA)

Advantages:

 Global address space provides a user-friendly programming perspective to memory

 Data sharing between tasks is both fast and uniform due to the proximity of memory to CPUs

Disadvantages:

 Primary disadvantage is the lack of scalability between memory and CPUs. Adding more CPUs
can geometrically increases traffic on the shared memory-CPU path, and for cache coherent
systems, geometrically increase traffic associated with cache/memory management.
 Programmer responsibility for synchronization constructs that ensure "correct" access of global
memory.

Parallel Computer Memory Architectures

Distributed Memory
General Characteristics:

 Like shared memory systems, distributed memory systems vary widely but share a common
characteristic. Distributed memory systems require a communication network to connect inter-
processor memory.
 28

 Processors have their own local memory. Memory addresses in one processor do not map to
another processor, so there is no concept of global address space across all processors.
 Because each processor has its own local memory, it operates independently. Changes it
makes to its local memory have no effect on the memory of other processors. Hence, the
concept of cache coherency does not apply.
 When a processor needs access to data in another processor, it is usually the task of the
programmer to explicitly define how and when data is communicated. Synchronization between
tasks is likewise the programmer's responsibility.
 The network "fabric" used for data transfer varies widely, though it can be as simple as Ethernet.

Advantages:

 Memory is scalable with the number of processors. Increase the number of processors and the
size of memory increases proportionately.
 Each processor can rapidly access its own memory without interference and without the
overhead incurred with trying to maintain global cache coherency.
 Cost effectiveness: can use commodity, off-the-shelf processors and networking.

Disadvantages:

 The programmer is responsible for many of the details associated with data communication
between processors.
 It may be difficult to map existing data structures, based on global memory, to this memory
organization.
 Non-uniform memory access times - data residing on a remote node takes longer to access than
node local data.
 29

Parallel Computer Memory Architectures

Hybrid Distributed-Shared Memory

General Characteristics:

 The largest and fastest computers in the world today employ both shared and distributed
memory architectures.

 The shared memory component can be a shared memory machine and/or graphics processing
units (GPU).
 The distributed memory component is the networking of multiple shared memory/GPU
machines, which know only about their own memory - not the memory on another machine.
Therefore, network communications are required to move data from one machine to another.
 Current trends seem to indicate that this type of memory architecture will continue to prevail and
increase at the high end of computing for the foreseeable future.

Advantages and Disadvantages:

 Whatever is common to both shared and distributed memory architectures.

 Increased scalability is an important advantage
 Increased programmer complexity is an important disadvantage
 30

Parallel Programming Models

Overview
 There are several parallel programming models in common use:
o Shared Memory (without threads)
o Threads
o Distributed Memory / Message Passing
o Data Parallel
o Hybrid
o Single Program Multiple Data (SPMD)
o Multiple Program Multiple Data (MPMD)
 Parallel programming models exist as an abstraction above hardware and memory
architectures.
 Although it might not seem apparent, these models are NOT specific to a particular type of
machine or memory architecture. In fact, any of these models can (theoretically) be implemented
on any underlying hardware. Two examples from the past are discussed below.

SHARED memory model on a DISTRIBUTED memory machine:

Kendall Square Research (KSR) ALLCACHE approach. Machine memory was physically distributed
across networked machines, but appeared to the user as a single shared memory global address space.
Generically, this approach is referred to as "virtual shared memory".

DISTRIBUTED memory model on a SHARED memory machine:

Message Passing Interface (MPI) on SGI Origin 2000. The SGI Origin 2000 employed the CC-NUMA
type of shared memory architecture, where every task has direct access to global address space spread
across all machines. However, the ability to send and receive messages using MPI, as is commonly done
over a network of distributed memory machines, was implemented and commonly used.
 31

 Which model to use? This is often a combination of what is available and personal choice.
There is no "best" model, although there certainly are better implementations of some models
over others.
 The following sections describe each of the models mentioned above, and also discuss some of
their actual implementations.

Shared Memory Model (without threads)

 In this programming model,
processes/tasks share a common address
space, which they read and write to
asynchronously.
 Various mechanisms such as locks /
semaphores are used to control access to
the shared memory, resolve contentions
and to prevent race conditions and
deadlocks.
 This is perhaps the simplest parallel
programming model.
 An advantage of this model from the
programmer's point of view is that the
notion of data "ownership" is lacking, so
there is no need to specify explicitly the
communication of data between tasks. All
processes see and have equal access to
shared memory. Program development can
often be simplified.
 An important disadvantage in terms of performance is that it becomes more difficult to
understand and manage data locality:
o Keeping data local to the process that works on it conserves memory accesses, cache
refreshes and bus traffic that occurs when multiple processes use the same data.
o Unfortunately, controlling data locality is hard to understand and may be beyond the
control of the average user.

Implementations:

 On stand-alone shared memory machines, native operating systems, compilers and/or hardware
provide support for shared memory programming. For example, the POSIX standard provides an
API for using shared memory, and UNIX provides shared memory segments (shmget, shmat,
shmctl, etc).
 On distributed memory machines, memory is physically distributed across a network of
machines, but made global through specialized hardware and software. A variety of SHMEM
implementations are available: http://en.wikipedia.org/wiki/SHMEM.

 32

Threads Model
 This programming model is a type of shared memory programming.
 In the threads model of parallel programming, a single "heavy weight" process can have multiple
"light weight", concurrent execution paths.
 For example:
o The main program a.out is
scheduled to run by the native
operating system. a.out loads and
acquires all of the necessary system
and user resources to run. This is
the "heavy weight" process.
o a.out performs some serial work,
and then creates a number of tasks
(threads) that can be scheduled and
run by the operating system
concurrently.
o Each thread has local data, but also,
shares the entire resources of a.out.
This saves the overhead associated
with replicating a program's
resources for each thread ("light
weight"). Each thread also benefits
from a global memory view because
it shares the memory space of a.out.
o A thread's work may best be
described as a subroutine within the
main program. Any thread can
execute any subroutine at the same
time as other threads.
o Threads communicate with each
other through global memory
(updating address locations). This
requires synchronization constructs
to ensure that more than one thread
is not updating the same global address at any time.
o Threads can come and go, but a.out remains present to provide the necessary shared
resources until the application has completed.

Implementations:

 From a programming perspective, threads implementations commonly comprise:

o A library of subroutines that are called from within parallel source code
o A set of compiler directives imbedded in either serial or parallel source code

In both cases, the programmer is responsible for determining the parallelism (although compilers
can sometimes help).
 33

 Threaded implementations are not new in computing. Historically, hardware vendors have
implemented their own proprietary versions of threads. These implementations differed
substantially from each other making it difficult for programmers to develop portable threaded
applications.
 Unrelated standardization efforts have resulted in two very different implementations of threads:
POSIX Threads and OpenMP.
 POSIX Threads
o Specified by the IEEE POSIX 1003.1c standard (1995). C Language only.
o Part of Unix/Linux operating systems
o Library based
o Commonly referred to as Pthreads.
o Very explicit parallelism; requires significant programmer attention to detail.
 OpenMP
o Industry standard, jointly defined and endorsed by a group of major computer hardware
and software vendors, organizations and individuals.
o Compiler directive based
o Portable / multi-platform, including Unix and Windows platforms
o Available in C/C++ and Fortran implementations
o Can be very easy and simple to use - provides for "incremental parallelism". Can begin
with serial code.
 Other threaded implementations exist, such as Microsoft's

More Information:

 POSIX Threads tutorial: computing.llnl.gov/tutorials/pthreads

 OpenMP tutorial: computing.llnl.gov/tutorials/openMP

Distributed Memory /
Message Passing
Model
 This model demonstrates the
following characteristics:
o A set of tasks that use
their own local memory
during computation.
Multiple tasks can reside
on the same physical
machine and/or across
an arbitrary number of
machines.
 34

o Tasks exchange data through communications by sending and receiving messages.

o Data transfer usually requires cooperative operations to be performed by each process.
For example, a send operation must have a matching receive operation.

Implementations:

 From a programming perspective, message passing implementations usually comprise a library

of subroutines. Calls to these subroutines are imbedded in source code. The programmer is
responsible for determining all parallelism.
 Historically, a variety of message passing libraries have been available since the 1980s. These
implementations differed substantially from each other making it difficult for programmers to
develop portable applications.
 In 1992, the MPI Forum was formed with the primary goal of establishing a standard interface for
message passing implementations.
 Part 1 of the Message Passing Interface (MPI) was released in 1994. Part 2 (MPI-2) was
released in 1996 and MPI-3 in 2012. All MPI specifications are available on the web at
http://www.mpi-forum.org/docs/.
 MPI is the "de facto" industry standard for message passing, replacing virtually all other
message passing implementations used for production work. MPI implementations exist for
virtually all popular parallel computing platforms. Not all implementations include everything in
MPI-1, MPI-2 or MPI-3.

Data Parallel Model

 May also be referred to as the
Partitioned Global Address Space
(PGAS) model.
 The data parallel model
demonstrates the following
characteristics:
o Address space is treated
globally
o Most of the parallel work
focuses on performing
operations on a data set. The
data set is typically organized
into a common structure,
such as an array or cube.
o A set of tasks work
collectively on the same data
structure, however, each task
works on a different partition
of the same data structure.
o Tasks perform the same
operation on their partition of
 35

work, for example, "add 4 to every array element".

 On shared memory architectures, all tasks may have access to the data structure through global
memory.
 On distributed memory architectures the data structure is split up and resides as "chunks" in the
local memory of each task.

Implementations:

 Currently, there are several relatively popular, and sometimes developmental, parallel
programming implementations based on the Data Parallel / PGAS model.
 Coarray Fortran: a small set of extensions to Fortran 95 for SPMD parallel programming.
Compiler dependent. More information: https://en.wikipedia.org/wiki/Coarray_Fortran
 Unified Parallel C (UPC): an extension to the C programming language for SPMD parallel
programming. Compiler dependent. More information: http://upc.lbl.gov/
 Global Arrays: provides a shared memory style programming environment in the context of
distributed array data structures. Public domain library with C and Fortran77 bindings. More
information: https://en.wikipedia.org/wiki/Global_Arrays
 X10: a PGAS based parallel programming language being developed by IBM at the Thomas J.
Watson Research Center. More information: http://x10-lang.org/
 Chapel: an open source parallel programming language project being led by Cray. More
information: http://chapel.cray.com/

Hybrid Model
 A hybrid model combines more than one of the previously described programming models.
 Currently, a common example of a hybrid model is the combination of the message passing model (MPI)
with the threads model (OpenMP).
o Threads perform computationally intensive kernels using local, on-node data
o Communications between processes on different nodes occurs over the network using MPI
 This hybrid model lends itself well to the most popular (currently) hardware environment of clustered
multi/many-core machines.
 36

 Another similar and increasingly popular example of a hybrid model is using MPI with CPU-GPU
(Graphics Processing Unit) programming.
o MPI tasks run on CPUs using local memory and communicating with each other over a network.
o Computationally intensive kernels are off-loaded to GPUs on-node.
o Data exchange between node-local memory and GPUs uses CUDA (or something equivalent).
 Other hybrid models are common:
o MPI with Pthreads
o MPI with non-GPU accelerators
o ...

SPMD and MPMD

Single Program Multiple Data (SPMD):

 SPMD is actually a "high level"

programming model that can be built
upon any combination of the
previously mentioned parallel
programming models.
 SINGLE PROGRAM: All tasks execute
their copy of the same program
simultaneously. This program can be
threads, message passing, data parallel or hybrid.
 MULTIPLE DATA: All tasks may use different data
 SPMD programs usually have the necessary logic programmed into them to allow different tasks
to branch or conditionally execute only those parts of the program they are designed to execute.
That is, tasks do not necessarily have to execute the entire program - perhaps only a portion of
it.
 The SPMD model, using message passing or hybrid programming, is probably the most
commonly used parallel programming model for multi-node clusters.

Multiple Program Multiple Data (MPMD):

 Like SPMD, MPMD is actually a "high

level" programming model that can be
built upon any combination of the
previously mentioned parallel
programming models.
 MULTIPLE PROGRAM: Tasks may
execute different programs
simultaneously. The programs can be
threads, message passing, data parallel or hybrid.
 MULTIPLE DATA: All tasks may use different data
 MPMD applications are not as common as SPMD applications, but may be better suited for
certain types of problems, particularly those that lend themselves better to functional
decomposition than domain decomposition (discussed later under Partioning).
 37

Designing Parallel Programs

Automatic vs. Manual Parallelization

 Designing and developing parallel programs has characteristically been a very manual process.
The programmer is typically responsible for both identifying and actually implementing
parallelism.
 Very often, manually developing parallel codes is a time consuming, complex, error-prone and
iterative process.
 For a number of years now, various tools have been available to assist the programmer with
converting serial programs into parallel programs. The most common type of tool used to
automatically parallelize a serial program is a parallelizing compiler or pre-processor.
 A parallelizing compiler generally works in two different ways:
o Fully Automatic
o The compiler analyzes the source code and identifies opportunities for parallelism.
o The analysis includes identifying inhibitors to parallelism and possibly a cost
weighting on whether or not the parallelism would actually improve performance.
o Loops (do, for) are the most frequent target for automatic parallelization.
o Programmer Directed
o Using "compiler directives" or possibly compiler flags, the programmer explicitly
tells the compiler how to parallelize the code.
o May be able to be used in conjunction with some degree of automatic
parallelization also.
 The most common compiler generated parallelization is done using on-node shared memory
and threads (such as OpenMP).
 If you are beginning with an existing serial code and have time or budget constraints, then
automatic parallelization may be the answer. However, there are several important caveats that
apply to automatic parallelization:
o Wrong results may be produced
o Performance may actually degrade
o Much less flexible than manual parallelization
o Limited to a subset (mostly loops) of code
o May actually not parallelize code if the compiler analysis suggests there are inhibitors or
the code is too complex
 The remainder of this section applies to the manual method of developing parallel codes.

Understand the Problem and the Program

 Undoubtedly, the first step in developing parallel software is to first understand the problem that
you wish to solve in parallel. If you are starting with a serial program, this necessitates
understanding the existing code also.
 Before spending time in an attempt to develop a parallel solution for a problem, determine
whether or not the problem is one that can actually be parallelized.
o Example of an easy to parallelize problem:
 38

Calculate the potential energy for each of several thousand independent

conformations of a molecule. When done, find the minimum energy
conformation.

o This problem is able to be solved in parallel. Each of the molecular conformations is

independently determinable. The calculation of the minimum energy conformation is also
a parallelizable problem.
o Example of a problem with little-to-no parallelism:

Calculation of the Fibonacci series (0,1,1,2,3,5,8,13,21,...) by use of the formula:

F(n) = F(n-1) + F(n-2)

o The calculation of the F(n) value uses those of both F(n-1) and F(n-2), which must be
computed first.

 Identify the program's hotspots:

o Know where most of the real work is being done. The majority of scientific and technical
programs usually accomplish most of their work in a few places.
o Profilers and performance analysis tools can help here
o Focus on parallelizing the hotspots and ignore those sections of the program that account for little
CPU usage.
 Identify bottlenecks in the program:
o Are there areas that are disproportionately slow, or cause parallelizable work to halt or be
deferred? For example, I/O is usually something that slows a program down.
o May be possible to restructure the program or use a different algorithm to reduce or eliminate
unnecessary slow areas
 Identify inhibitors to parallelism. One common class of inhibitor is data dependence, as demonstrated by
the Fibonacci sequence above.
 Investigate other algorithms if possible. This may be the single most important consideration when
designing a parallel application.
 Take advantage of optimized third party parallel software and highly optimized math libraries available
from leading vendors (IBM's ESSL, Intel's MKL, AMD's AMCL, etc.).

Partitioning
 One of the first steps in designing a parallel program is to break the problem into discrete
"chunks" of work that can be distributed to multiple tasks. This is known as decomposition or
partitioning.
 There are two basic ways to partition computational work among parallel tasks: domain
decomposition and functional decomposition.
 39

Domain Decomposition:

 In this type of partitioning, the data associated with a problem is decomposed. Each parallel task
then works on a portion of the data.

 There are different ways to partition data:

Functional Decomposition:

 In this approach, the focus is on the computation that is to be performed rather than on the data
manipulated by the computation. The problem is decomposed according to the work that must
be done. Each task then performs a portion of the overall work.
 40

 Functional decomposition lends itself well to problems that can be split into different tasks. For
example:

Ecosystem Modeling
Each program calculates the population of a given group, where each group's growth depends
on that of its neighbors. As time progresses, each process calculates its current state, then
exchanges information with the neighbor populations. All tasks then progress to calculate the
state at the next time step.

Signal Processing
An audio signal data set is passed through four distinct computational filters. Each filter is a
separate process. The first segment of data must pass through the first filter before progressing
to the second. When it does, the second segment of data passes through the first filter. By the
time the fourth segment of data is in the first filter, all four tasks are busy.
 41

Climate Modeling
Each model component can be thought of as a separate task. Arrows represent exchanges of
data between components during computation: the atmosphere model generates wind velocity
data that are used by the ocean model, the ocean model generates sea surface temperature
data that are used by the atmosphere model, and so on.

 Combining these two types of problem decomposition is common and natural.

Communications
Who Needs Communications?

The need for communications between tasks depends upon your problem:

 You DON'T need communications

o Some types of problems can be decomposed and executed in parallel with virtually no
need for tasks to share data. For example, imagine an image processing operation where
every pixel in a black and white image needs to have its color reversed. The image data
can easily be distributed to multiple tasks that then act independently of each other to do
their portion of the work.
o These types of problems are often called embarrassingly parallel because they are so
straight-forward. Very little inter-task communication is required.

 You DO need communications

o Most parallel applications are not quite so simple, and do require tasks to share data with
each other. For example, a 3-D heat diffusion problem requires a task to know the
temperatures calculated by the tasks that have neighboring data. Changes to neighboring
data has a direct effect on that task's data.
 42

Factors to Consider:

There are a number of important factors to consider when designing your program's inter-task
communications:

 Cost of communications
o Inter-task communication virtually always implies overhead.
o Machine cycles and resources that could be used for computation are instead used to
package and transmit data.
o Communications frequently require some type of synchronization between tasks, which
can result in tasks spending time "waiting" instead of doing work.
o Competing communication traffic can saturate the available network bandwidth, further
aggravating performance problems.
 Latency vs. Bandwidth
o latency is the time it takes to send a minimal (0 byte) message from point A to point B.
Commonly expressed as microseconds.
o bandwidth is the amount of data that can be communicated per unit of time. Commonly
expressed as megabytes/sec or gigabytes/sec.
o Sending many small messages can cause latency to dominate communication
overheads. Often it is more efficient to package small messages into a larger message,
thus increasing the effective communications bandwidth.
 Visibility of communications
o With the Message Passing Model, communications are explicit and generally quite visible
and under the control of the programmer.
o With the Data Parallel Model, communications often occur transparently to the
programmer, particularly on distributed memory architectures. The programmer may not
even be able to know exactly how inter-task communications are being accomplished.
 Synchronous vs. asynchronous communications
o Synchronous communications require some type of "handshaking" between tasks that are
sharing data. This can be explicitly structured in code by the programmer, or it may
happen at a lower level unknown to the programmer.
o Synchronous communications are often referred to as blocking communications since
other work must wait until the communications have completed.
o Asynchronous communications allow tasks to transfer data independently from one
another. For example, task 1 can prepare and send a message to task 2, and then
immediately begin doing other work. When task 2 actually receives the data doesn't
matter.
o Asynchronous communications are often referred to as non-blocking communications
since other work can be done while the communications are taking place.
o Interleaving computation with communication is the single greatest benefit for using
asynchronous communications.
 Scope of communications
o Knowing which tasks must communicate with each other is critical during the design
stage of a parallel code. Both of the two scopings described below can be implemented
synchronously or asynchronously.
o Point-to-point - involves two tasks with one task acting as the sender/producer of data,
and the other acting as the receiver/consumer.
 43

o Collective - involves data sharing between more than two tasks, which are often
specified as being members in a common group, or collective. Some common variations
(there are more):

 Efficiency of communications
o Very often, the programmer will have a choice with regard to factors that can affect
communications performance. Only a few are mentioned here.
o Which implementation for a given model should be used? Using the Message Passing
Model as an example, one MPI implementation may be faster on a given hardware
platform than another.
o What type of communication operations should be used? As mentioned previously,
asynchronous communication operations can improve overall program performance.
o Network media - some platforms may offer more than one network for communications.
Which one is best?
 44

 Overhead and Complexity

 Finally, realize that this is only a partial list of things to consider!!!

Synchronization
 Managing the sequence of work and the tasks
performing it is a critical design consideration for
most parallel programs.
 Can be a significant factor in program performance
(or lack of it)
 Often requires "serialization" of segments of the
program.

Types of Synchronization:

 Barrier
o Usually implies that all tasks are involved
o Each task performs its work until it reaches the barrier. It then stops, or "blocks".
o When the last task reaches the barrier, all tasks are synchronized.
o What happens from here varies. Often, a serial section of work must be done. In other
cases, the tasks are automatically released to continue their work.
 Lock / semaphore
o Can involve any number of tasks
o Typically used to serialize (protect) access to global data or a section of code. Only one
task at a time may use (own) the lock / semaphore / flag.
 45

o The first task to acquire the lock "sets" it. This task can then safely (serially) access the
protected data or code.
o Other tasks can attempt to acquire the lock but must wait until the task that owns the lock
releases it.
o Can be blocking or non-blocking
 Synchronous communication operations
o Involves only those tasks executing a communication operation
o When a task performs a communication operation, some form of coordination is required
with the other task(s) participating in the communication. For example, before a task can
perform a send operation, it must first receive an acknowledgment from the receiving task
that it is OK to send.
o Discussed previously in the Communications section.

Data Dependencies
Definition:

 A dependence exists between program

statements when the order of statement
execution affects the results of the program.
 A data dependence results from multiple use
of the same location(s) in storage by different
tasks.
 Dependencies are important to parallel
programming because they are one of the
primary inhibitors to parallelism.

Examples:

 Loop carried data dependence

DO 500 J = MYSTART,MYEND
A(J) = A(J-1) * 2.0
500 CONTINUE

The value of A(J-1) must be computed before the value of A(J), therefore A(J) exhibits a data
dependency on A(J-1). Parallelism is inhibited.

If Task 2 has A(J) and task 1 has A(J-1), computing the correct value of A(J) necessitates:
 46

o Distributed memory architecture - task 2 must obtain the value of A(J-1) from task 1 after
task 1 finishes its computation
o Shared memory architecture - task 2 must read A(J-1) after task 1 updates it

 Loop independent data dependence

task 1 task 2
------ ------

X = 2 X = 4
. .
. .
Y = X**2 Y = X**3

As with the previous example, parallelism is inhibited. The value of Y is dependent on:

o Distributed memory architecture - if or when the value of X is communicated between the

tasks.
o Shared memory architecture - which task last stores the value of X.
 Although all data dependencies are important to identify when designing parallel programs, loop
carried dependencies are particularly important since loops are possibly the most common
target of parallelization efforts.

How to Handle Data Dependencies:

 Distributed memory architectures - communicate required data at synchronization points.

 Shared memory architectures -synchronize read/write operations between tasks.

Load Balancing
 Load balancing refers to the practice of distributing approximately equal amounts of work among
tasks so that all tasks are kept busy all of the time. It can be considered a minimization of task
idle time.
 Load balancing is important to parallel programs for performance reasons. For example, if all
tasks are subject to a barrier synchronization point, the slowest task will determine the overall
performance.
 47

How to Achieve Load Balance:

 Equally partition the work each task receives

o For array/matrix operations where each task performs similar work, evenly distribute the
data set among the tasks.
o For loop iterations where the work done in each iteration is similar, evenly distribute the
iterations across the tasks.
o If a heterogeneous mix of machines with varying performance characteristics are being
used, be sure to use some type of performance analysis tool to detect any load
imbalances. Adjust work accordingly.

 Use dynamic work assignment

o Certain classes of problems result in load imbalances even if data is evenly distributed
among tasks:
o Sparse arrays - some tasks will have actual data to work on while others have
mostly "zeros".
o Adaptive grid methods - some tasks may need to refine their mesh while others
don't.
o N-body simulations - where some particles may migrate to/from their original task
domain to another task's; where the particles owned by some tasks require more
work than those owned by other tasks.
o When the amount of work each task will perform is intentionally variable, or is unable to
be predicted, it may be helpful to use a scheduler - task pool approach. As each task
finishes its work, it queues to get a new piece of work.
o It may become necessary to design an algorithm which detects and handles load
imbalances as they occur dynamically within the code.

Granularity
Computation / Communication Ratio:

 In parallel computing, granularity is a qualitative measure of the ratio of computation to

communication.
 48

 Periods of computation are typically separated from periods of communication by

synchronization events.

Fine-grain Parallelism:

 Relatively small amounts of computational work are done

between communication events
 Low computation to communication ratio
 Facilitates load balancing
 Implies high communication overhead and less
opportunity for performance enhancement
 If granularity is too fine it is possible that the overhead
required for communications and synchronization between
tasks takes longer than the computation.

Coarse-grain Parallelism:

 Relatively large amounts of computational work are done

between communication/synchronization events
 High computation to communication ratio
 Implies more opportunity for performance increase
 Harder to load balance efficiently

Which is Best?

 The most efficient granularity is dependent on the

algorithm and the hardware environment in which it runs.
 In most cases the overhead associated with
communications and synchronization is high relative to
execution speed so it is advantageous to have coarse
granularity.
 Fine-grain parallelism can help reduce overheads due to
load imbalance.

I/O
The Bad News:

 I/O operations are generally regarded as inhibitors to parallelism.

 I/O operations require orders of magnitude more time than memory operations.
 Parallel I/O systems may be immature or not available for all platforms.
 In an environment where all tasks see the same file space, write operations can result in file
overwriting.
 Read operations can be affected by the file server's ability to handle multiple read requests at
the same time.
 49

 I/O that must be conducted over the network (NFS, non-local) can cause severe bottlenecks and
even crash file servers.

The Good News:

 Parallel file systems are available. For example:

o GPFS: General Parallel File System (IBM)
o Lustre: for Linux clusters (Intel)
o OrangeFS: Open source parallel file system follow on to Parallel Virtual File System
(PVFS)
o PanFS: Panasas ActiveScale File System for Linux clusters (Panasas, Inc.)
o And more - see
http://en.wikipedia.org/wiki/List_of_file_systems#Distributed_parallel_file_systems
 The parallel I/O programming interface specification for MPI has been available since 1996 as
part of MPI-2. Vendor and "free" implementations are now commonly available.
 A few pointers:
o Rule #1: Reduce overall I/O as much as possible
o If you have access to a parallel file system, use it.
o Writing large chunks of data rather than small chunks is usually significantly more
efficient.
o Fewer, larger files performs better than many small files.
o Confine I/O to specific serial portions of the job, and then use parallel communications to
distribute data to parallel tasks. For example, Task 1 could read an input file and then
communicate required data to other tasks. Likewise, Task 1 could perform write operation
after receiving required data from all other tasks.
o Aggregate I/O operations across tasks - rather than having many tasks perform I/O, have
a subset of tasks perform it.

Debugging
 Debugging parallel codes can be incredibly difficult, particularly as codes scale upwards.
 The good news is that there are some excellent debuggers available to assist:
o Threaded - pthreads and OpenMP
 50

o MPI
o GPU / accelerator
o Hybrid
 Livermore Computing users have access to several parallel debugging tools installed on LC's
clusters:
o TotalView from RogueWave Software
o DDT from Allinea
o Inspector from Intel
o Stack Trace Analysis Tool (STAT) - locally developed
 All of these tools have a learning curve associated with them - some more than others.
 For details and getting started information, see:
o LC's "Supported Software and Computing Tools" web pages at
https://computing.llnl.gov/?set=code&page=software_tools
o TotalView tutorial: https://computing.llnl.gov/tutorials/totalview/

Performance Analysis and Tuning

 As with debugging, analyzing and tuning parallel program performance can be much more
challenging than for serial programs.
 Fortunately, there are a number of excellent tools for parallel program performance analysis and
tuning.
 Livermore Computing users have access to several such tools, most of which are available on
all production clusters.
 Some starting points for tools installed on LC systems:
o LC's "Supported Software and Computing Tools" web pages at
https://computing.llnl.gov/?set=code&page=software_tools
o TAU: http://www.cs.uoregon.edu/research/tau/docs.php
o HPCToolkit: http://hpctoolkit.org/documentation.html
o Open|Speedshop: http://www.openspeedshop.org/
o Vampir / Vampirtrace: http://vampir.eu/
o Valgrind: http://valgrind.org/
o PAPI: http://icl.cs.utk.edu/papi/
 51

o mpitrace https://computing.llnl.gov/tutorials/bgq/index.html#mpitrace
o mpiP: http://mpip.sourceforge.net/
o memP: http://memp.sourceforge.net/
 52

Parallel Examples

Array Processing
 This example demonstrates calculations on 2-dimensional
array elements; a function is evaluated on each array
element.
 The computation on each array element is independent
from other array elements.
 The problem is computationally intensive.
 The serial program calculates one element at a time in
sequential order.
 Serial code could be of the form:

do j = 1,n
do i = 1,n
a(i,j) = fcn(i,j)
end do
end do

 Questions to ask:
o Is this problem able to be parallelized?
o How would the problem be partitioned?
o Are communications needed?
o Are there any data dependencies?
o Are there synchronization needs?
o Will load balancing be a concern?

Array Processing
Parallel Solution 1
 The calculation of elements is independent of one
another - leads to an embarrassingly parallel
solution.
 Arrays elements are evenly distributed so that each
process owns a portion of the array (subarray).
o Distribution scheme is chosen for efficient
memory access; e.g. unit stride (stride of 1)
through the subarrays. Unit stride
maximizes cache/memory usage.
 53

o Since it is desirable to have unit stride through the subarrays, the choice of a distribution
scheme depends on the programming language. See the Block - Cyclic Distributions
Diagram for the options.
 Independent calculation of array elements ensures there is no need for communication or
synchronization between tasks.
 Since the amount of work is evenly distributed across processes, there should not be load
balance concerns.
 After the array is distributed, each task executes the portion of the loop corresponding to the
data it owns.
For example, both Fortran (column-major) and C (row-major) block distributions are shown:

do j = mystart, myend for i (i = mystart; i < myend;

do i = 1, n i++) {
a(i,j) = fcn(i,j) for j (j = 0; j < n; j++) {
end do a(i,j) = fcn(i,j);
end do }
}

 Notice that only the outer loop variables are different from the serial solution.

One Possible Solution:

 Implement as a Single Program Multiple Data (SPMD) model - every task executes the same
program.
 Master process initializes array, sends info to worker processes and receives results.
 Worker process receives info, performs its share of computation and sends results to master.
 Using the Fortran storage scheme, perform block distribution of the array.
 Pseudo code solution: red highlights changes for parallelism.

find out if I am MASTER or WORKER

if I am MASTER

initialize the array

send each WORKER info on part of array it owns
send each WORKER its portion of initial array

receive from each WORKER results

else if I am WORKER
receive from MASTER info on part of array I own
receive from MASTER my portion of initial array
 54

# calculate my portion of array

do j = my first column,my last column
do i = 1,n
a(i,j) = fcn(i,j)
end do
end do

send MASTER results

endif

Example Programs:

 MPI Program in C:
 MPI Program in Fortran:

Array Processing
Parallel Solution 2: Pool of Tasks
 The previous array solution demonstrated static load balancing:
o Each task has a fixed amount of work to do
o May be significant idle time for faster or more lightly loaded processors - slowest tasks
determines overall performance.
 Static load balancing is not usually a major concern if all tasks are performing the same amount
of work on identical machines.
 If you have a load balance problem (some tasks work faster than others), you may benefit by
using a "pool of tasks" scheme.

Pool of Tasks Scheme:

 Two processes are employed

Master Process:

o Holds pool of tasks for worker processes to do

o Sends worker a task when requested
o Collects results from workers

Worker Process: repeatedly does the following

o Gets task from master process

o Performs computation
o Sends results to master
 55

 Worker processes do not know before runtime which portion of array they will handle or how
many tasks they will perform.
 Dynamic load balancing occurs at run time: the faster tasks will get more work to do.
 Pseudo code solution: red highlights changes for parallelism.

find out if I am MASTER or WORKER

if I am MASTER

do until no more jobs

if request send to WORKER next job
else receive results from WORKER
end do

else if I am WORKER

do until no more jobs

request job from MASTER
receive from MASTER next job

calculate array element: a(i,j) = fcn(i,j)

send results to MASTER

end do

endif

Discussion:

 In the above pool of tasks example, each task calculated an individual array element as a job.
The computation to communication ratio is finely granular.
 Finely granular solutions incur more communication overhead in order to reduce task idle time.
 A more optimal solution might be to distribute more work with each job. The "right" amount of
work is problem dependent.

Parallel Examples

PI Calculation
 56

 The value of PI can be calculated in a number

of ways. Consider the following method of
approximating PI
1. Inscribe a circle in a square
2. Randomly generate points in the
square
3. Determine the number of points in the
square that are also in the circle
4. Let r be the number of points in the
circle divided by the number of points
in the square
5. PI ~ 4 r
6. Note that the more points generated,
the better the approximation
 Serial pseudo code for this procedure:

npoints = 10000
circle_count = 0

do j = 1,npoints
generate 2 random numbers
between 0 and 1
xcoordinate = random1
ycoordinate = random2
if (xcoordinate,
ycoordinate) inside circle
then circle_count =
circle_count + 1
end do

PI = 4.0*circle_count/npoints

 The problem is computationally intensive -

most of the time is spent executing the loop
 Questions to ask:
o Is this problem able to be parallelized?
o How would the problem be
partitioned?
o Are communications needed?
o Are there any data dependencies?
o Are there synchronization needs?
o Will load balancing be a concern?
 57

PI Calculation
Parallel Solution
 Another problem that's easy to parallelize:
o All point calculations are independent; no data dependencies
o Work can be evenly divided; no load balance concerns
o No need for communication or synchronization between tasks
 Parallel strategy:
o Divide the loop into equal portions that can be executed by the pool of tasks
o Each task independently performs its work
o A SPMD model is used
o One task acts as the master to collect results and compute the value of PI
 Pseudo code solution: red highlights changes for parallelism.

npoints = 10000
circle_count = 0

p = number of tasks
num = npoints/p
 58

find out if I am MASTER or WORKER

do j = 1,num
generate 2 random numbers between 0 and 1
xcoordinate = random1
ycoordinate = random2
if (xcoordinate, ycoordinate) inside circle
then circle_count = circle_count + 1
end do

if I am MASTER

receive from WORKERS their circle_counts

compute PI (use MASTER and WORKER calculations)

else if I am WORKER

send to MASTER circle_count

endif

Parallel Examples

Simple Heat Equation

 Most problems in parallel computing require
communication among the tasks. A number of
common problems require communication with
"neighbor" tasks.
 The heat equation describes the temperature change
over time, given initial temperature distribution and
boundary conditions.
 A finite differencing scheme is employed to solve the
heat equation numerically on a square region.
o The elements of a 2-dimensional array
represent the temperature at points on the
square.
o The initial temperature is zero on the
boundaries and high in the middle.
o The boundary temperature is held at zero.
o A time stepping algorithm is used.
 The calculation of an element is dependent upon
 59

neighbor element values:

 A serial program would contain code like:

do iy = 2, ny - 1
do ix = 2, nx - 1
u2(ix, iy) = u1(ix, iy) +
cx * (u1(ix+1,iy) + u1(ix-
1,iy) - 2.*u1(ix,iy)) +
cy * (u1(ix,iy+1) +
u1(ix,iy-1) - 2.*u1(ix,iy))
end do
end do

 Questions to ask:
o Is this problem able to be parallelized?
o How would the problem be partitioned?
o Are communications needed?
o Are there any data dependencies?
o Are there synchronization needs?
o Will load balancing be a concern?

Simple Heat Equation

Parallel Solution
 This problem is more challenging, since there data
dependencies, which require communications and
synchronization.
 The entire array is partitioned and distributed as
subarrays to all tasks. Each task owns an equal
portion of the total array.
 Because the amount of work is equal, load
balancing should not be a concern
 Determine data dependencies:
o interior elements belonging to a task are
independent of other tasks
 60

o border elements are dependent upon a neighbor task's data, necessitating

communication.
 Implement as an SPMD model:
o Master process sends initial info to workers, and then waits to collect results from all
workers
o Worker processes calculate solution within specified number of time steps,
communicating as necessary with neighbor processes
 Pseudo code solution: red highlights changes for parallelism.

find out if I am MASTER or WORKER

if I am MASTER
initialize array
send each WORKER starting info and subarray
receive results from each WORKER

else if I am WORKER
receive from MASTER starting info and subarray

# Perform time steps

do t = 1, nsteps
update time
send neighbors my border info
receive from neighbors their border info
update my portion of solution array

end do

send MASTER results

endif

Example Programs:

 MPI Program in C:
 MPI Program in Fortran:
 61

Parallel Examples

1-D Wave Equation

 In this example, the amplitude along a uniform, vibrating string is calculated after a specified
amount of time has elapsed.
 The calculation involves:
o the amplitude on the y axis
o i as the position index along the x axis
o node points imposed along the string
o update of the amplitude at discrete time steps.

 The equation to be solved is the one-dimensional wave equation:

A(i,t+1) = (2.0 * A(i,t)) - A(i,t-1) + (c * (A(i-1,t) - (2.0 * A(i,t))

+ A(i+1,t)))

 where c is a constant
 Note that amplitude will depend on previous timesteps (t, t-1) and neighboring points (i-1, i+1).
 Questions to ask:
o Is this problem able to be parallelized?
o How would the problem be partitioned?
o Are communications needed?
o Are there any data dependencies?
o Are there synchronization needs?
o Will load balancing be a concern?
 62

1-D Wave Equation

Parallel Solution
 This is another example of a problem involving data dependencies. A parallel solution will
involve communications and synchronization.
 The entire amplitude array is partitioned and distributed as subarrays to all tasks. Each task
owns an equal portion of the total array.
 Load balancing: all points require equal work, so the points should be divided equally
 A block decomposition would have the work partitioned into the number of tasks as chunks,
allowing each task to own mostly contiguous data points.
 Communication need only occur on data borders. The larger the block size the less the
communication.

 Implement as an SPMD model:

o Master process sends initial info to workers, and then waits to collect results from all
workers
o Worker processes calculate solution within specified number of time steps,
communicating as necessary with neighbor processes
 Pseudo code solution: red highlights changes for parallelism.

find out number of tasks and task identities

#Identify left and right neighbors

left_neighbor = mytaskid - 1
right_neighbor = mytaskid +1
if mytaskid = first then left_neigbor = last
if mytaskid = last then right_neighbor = first

find out if I am MASTER or WORKER

if I am MASTER
initialize array
send each WORKER starting info and subarray
 63

else if I am WORKER`
receive starting info and subarray from MASTER
endif

#Perform time steps

#In this example the master participates in calculations
do t = 1, nsteps
send left endpoint to left neighbor
receive left endpoint from right neighbor
send right endpoint to right neighbor
receive right endpoint from left neighbor

#Update points along line

do i = 1, npoints
newval(i) = (2.0 * values(i)) - oldval(i)
+ (sqtau * (values(i-1) - (2.0 * values(i)) + values(i+1)))
end do

end do

#Collect results and write to file

if I am MASTER
receive results from each WORKER
write results to file
else if I am WORKER
send results to MASTER
endif
 64

References and More Information

 Author: Blaise Barney, Livermore Computing.
 A search on the WWW for "parallel programming" or "parallel computing" will yield a wide variety
of information.
 Recommended reading:
o "Designing and Building Parallel Programs". Ian Foster.
http://www.mcs.anl.gov/~itf/dbpp/
o "Introduction to Parallel Computing". Ananth Grama, Anshul Gupta, George Karypis, Vipin
Kumar.
http://www-users.cs.umn.edu/~karypis/parbook/
o "Overview of Recent Supercomputers". A.J. van der Steen, Jack Dongarra.
OverviewRecentSupercomputers.2008.pdf
 Photos/Graphics have been created by the author, created by other LLNL employees, obtained
from non-copyrighted, government or public domain (such as http://commons.wikimedia.org/)
sources, or used with the permission of authors from other presentations and web pages.
 History: These materials have evolved from the following sources, which are no longer
maintained or available.
o Tutorials located in the Maui High Performance Computing Center's "SP Parallel
Programming Workshop".
o Tutorials located at the Cornell Theory Center's "Education and Training" web page.

https://computing.llnl.gov/tutorials/parallel_comp/
Last Modified: 06/19/2016 18:31:26 [email protected]
UCRL-MI-133316