Dot net parallelism and multicore computing

DOT NET PARALLELISM & MULTICORE COMPUTING
BY
ARAVINDHAN G
OMNIEXTRACT TEAM
GENESIS GROUP

Why parallelism?
What is parallelism?
Types of parallelism in C#
Data Parallelism
Parallel Loops
Task Parallelism
2
AGENDA

 Don’t expect your sequential program to run faster on new
processors.
 Still, processor technology advances
 BUT the focus now is on multiple cores per chip
 Today’s desktops typically have 4 cores
 The multi-core revolution puts pressure on software developers
to use parallelism if they want to benefit from future hardware
improvements
4
The Free Lunch is over

 Parallelism means that a programming task can be split into
parts that can be run on several networked processors or
computers.
 Parallel computing is a form of computation in which many
calculations are carried out simultaneously, operating on the
principle that large problems can often be divided into smaller
ones, which are then solved concurrently ("in parallel").
5
What is parallelism?

C# supports two main models of parallelism:
• Data parallelism: where an operation is applied to each
element in a collection.
• Task parallelism: where independent computations are
executed in parallel.
6
Types of Parallelism in C#

A sequential for loop in C#:
int n = ...
for (int i = 0; i<=n; i++)
{
// ...
}
A parallel for loop in C#:
int n = ...
Parallel.For(0, n, i =>
{
// ...
});
7
Parallel Loops in C#

 The language construct for is translated into a
(higher-order) function Parallel.For.
 The argument to Parallel.For is an anonymous method,
specifying the code to be performed in each loop
iteration.
 The arguments to this anonymous method are the start
value, the end value and the iteration variable.
8
Parallel Loops in C#

We can limit the degree of parallelism like this:
var options = new ParallelOptions() {
MaxDegreeOfParallelism = 2 };
Parallel.For(0, n, options, i =>
{
fibs[i] = Fib(i);
});
9
A Simple Example

 Parallel loops have two ways to break or stop a loop
instead of just one.
 Parallel break, loopState.Break(), allows all steps with
indices lower than the break index to run before
terminating the loop.
 Parallel stop, loopState.Stop(), terminates the loop
without allowing any new steps to begin.
10
Terminating a Parallel Loop

 The parallel aggregate pattern combines data
parallelism over a collection, with the aggregation of the
result values to an overall result.
 It is parameterized both over the operation on each
element as well as the combination (aggregation) of the
partial results to an overall results.
An Example of Parallel Aggregates:
var options = new ParallelOptions() {
MaxDegreeOfParallelism = k};
Parallel.ForEach(seq /* sequence */, options,
() => 0, // The local initial partial result
// The loop body
});
11
Parallel Aggregates

 When independent computations are started in different
tasks, we use a model of task parallelism.
 This model is more general than data parallelism, but
requires more detailed control of synchronization and
communication.
 The most basic construct for task parallelism is:
Parallel.Invoke(DoLeft, DoRight);
 It executes the methods DoLeft and DoRight in parallel,
and waits for both of them to finish.
12
Task Parallelism in C#

The following code sorts 2 lists in parallel, providing a
comparison operation as an argument:
Parallel.Invoke( // generate two parallel threads
() => ic1.Sort(cmp_int_lt),
() => ic2.Sort(cmp_int_gt));
13
Example of Task Parallelism

 The implementation of Invoke uses the more basic
constructs
StartNew, for starting a computation;
Wait, WaitAll, WaitAny, for synchronising several
computations.
 Any shared data structure needs to be protected with
locks, semaphores or such.
 Programming on this level is similar to explicitly
managing threads:
it can be more efficient
it is error-prone.
14
Implementation of Task Parallelism

static void SequentialQuickSort(int[] array, int from, int to)
{
if (to - from <= Threshold)
{
InsertionSort(array, from, to);
}
else
{
int pivot = from + (to - from) / 2;
pivot = Partition(array, from, to, pivot);
SequentialQuickSort(array, from, pivot - 1);
SequentialQuickSort(array, pivot + 1, to);
}
}
15
Example: Sequential Code

static void ParallelQuickSort(int[] array, int from, int to, int depthRemaining)
{
if (to - from <= Threshold)
{
InsertionSort(array, from, to);
}
else
{
int pivot = from + (to - from) / 2;
pivot = Partition(array, from, to, pivot);
if (depthRemaining > 0)
{
Parallel.Invoke(
() => ParallelQuickSort(array, from, pivot - 1,
depthRemaining - 1),
() => ParallelQuickSort(array, pivot + 1, to,
depthRemaining - 1));
}
else
{
ParallelQuickSort(array, from, pivot - 1, 0);
ParallelQuickSort(array, pivot + 1, to, 0);
}
}
}
16
Example: Parallel Code

 The preferred, high-level way of coding parallel
computation in C# is through parallel patterns, an
instance of design patterns.
 Parallel patterns capture common patterns of parallel
computation.
 Two main classes of parallelism exist:
Data parallelism, which is implemented through parallel
For/Foreach loops.
Task parallelism, which is implemented through parallel
method invocation.
 Tuning the parallel performance often requires code
restructuring.
17
Summary

Dot net parallelism and multicore computing

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Dot net parallelism and multicore computing (20)

Recently uploaded (20)

Dot net parallelism and multicore computing