COA UNIT 5

0Parallelism is a concept where multiple tasks or operations are executed at the same time
to increase the speed and efficiency of computing. This is done by using multiple processors
or parts of the CPU (like ALUs, or Arithmetic Logic Units) to handle different tasks
simultaneously. Let’s break it down in simple terms and go over each aspect of parallelism
mentioned:
What is Parallelism?
 Basic Definition: When two or more operations are performed simultaneously, it's
called parallelism.
 Goal: The main aim is to make computing faster by handling multiple tasks at once,
rather than one at a time.
 Parallel Computers: These are systems with multiple processors that work together
to solve a big problem, effectively speeding up the overall process.
Goals of Parallelism
1. Faster Computation: Parallelism reduces the time it takes to solve complex problems
by splitting tasks across processors.
2. Increased Throughput: More processing is done within the same time frame, making
systems more efficient.
3. Better Performance: By performing multiple operations at once, computers can
achieve more with the same clock speed.
4. Solving Bigger Problems: Parallelism allows systems to handle tasks that would be
too large or slow for a single CPU.
Real-World Applications of Parallelism
Parallelism is widely used in applications that require large amounts of computation or data
processing. Examples include:
 Weather Forecasting: Complex models run faster using multiple processors.
 Socio-Economic Models: Parallel computing helps handle data for large populations.
 Finite Element Analysis: Engineering simulations can use parallelism for quicker and
more accurate results.
 Artificial Intelligence: AI tasks, like image processing, rely on parallelism to handle
complex calculations.
 Genetic Engineering: Processing genetic data requires a lot of computation, which
parallelism can handle effectively.
  Defense and Medical Applications: Parallelism enables complex simulations and
large-scale data analysis.
Types of Parallelism
Parallelism can be achieved through hardware or software.
1. Hardware Parallelism
 Objective: To increase the speed of processing by designing computers with multiple
processors, cores, or threads.
 Processor Parallelism: Multiple CPUs, cores, or threads work together. For example,
a multi-core processor can run different parts of a program on each core.
 Memory Parallelism: Shared or distributed memory configurations allow different
processors to access data simultaneously. This structure is helpful for handling large,
complex tasks.
 Pipelining: Overlapping or pipelining instructions, where one instruction starts
before the previous one finishes, helps achieve parallelism.
2. Software Parallelism
 Definition: Software parallelism depends on how a program is written, including how
instructions are ordered and how data flows within the program.
 Control and Data Dependence: Programs are analyzed to see which parts can run
independently (parallel) and which depend on other steps.
 Program Flow Graph: This graph shows which operations can be done
simultaneously and which need to wait, helping identify the degree of parallelism in
the software.
 Variable Parallelism: As a program runs, the level of parallelism can change
depending on the tasks being executed.
Hardware Parallelism Details
 Instruction Issuing: A processor can issue multiple instructions per cycle (e.g., a 2-
issue or 3-issue processor), which makes it capable of parallel processing.
 Multi-Issue Processors: If a system has multiple processors, each issuing multiple
instructions per cycle, it can handle more tasks at once, improving throughput and
performance.
Software Parallelism Details
 Program Structure: How a program is structured affects its parallelism. For instance,
a well-optimized program can have many instructions that can be executed
simultaneously.
 Execution Variation: Software parallelism isn't always consistent—some parts of a
program may run in parallel, while others must run sequentially, depending on
dependencies.




Software parallelism is all about running different parts of a program simultaneously, and it
can be achieved at various levels. Each level of parallelism has a distinct approach and
applications, so let's explore the different types in detail.

1. Instruction Level Parallelism (ILP)

 What It Is: ILP is the degree to which individual instructions within a single program
can be executed in parallel. It depends on finding and running independent
instructions that do not rely on each other’s outcomes.
 How It Works: The processor and compiler work together to identify instructions
that can run simultaneously or be reordered for more efficient execution.
 Example:
o Consider three instructions:
1. x = a + b
2. y = c - d
3. z = x * y
o Instructions 1 and 2 don’t depend on each other, so they can run
simultaneously. However, instruction 3 depends on the results of 1 and 2, so
it can only start once they’re finished.
 Benefits: By overlapping the execution of independent instructions, ILP increases the
speed of execution within a single processor. A high ILP indicates efficient use of the
CPU, achieving better throughput at the same clock rate.
  Superscalar Architecture: Some modern processors, known as superscalar
processors, implement ILP by issuing multiple instructions per clock cycle, allowing
for faster processing within the same CPU.


2. Data Level Parallelism (DLP)

 What It Is: DLP occurs when the same operation is applied to multiple data points
simultaneously, particularly in parallel computing environments.
 How It Works: Data is distributed across different processing units or nodes, each
performing the same operation on its portion of the data.
 Example:
o Suppose we want to add all elements in an array. In sequential execution, it
would take n * Ta time units for an array of n elements, where Ta is the time
for one addition.
o

o In a data-parallel setup with four processors, the time taken would reduce to
(n/4) * Ta plus a small merging overhead.
o Another example is matrix multiplication, where each processor can handle
different parts of the matrix, significantly reducing computation time.
 Locality of Data: DLP's performance is affected by data locality, which refers to how
data is accessed and managed in memory. When data is close in memory, it’s faster
to access, especially with cache usage.
 Applications: Common in tasks that handle large data sets, such as scientific
computing, image processing, and machine learning.

3. Task Level Parallelism (TLP)

 What It Is: TLP divides a program into distinct tasks or functions that can run
independently on separate processors.
 How It Works: Tasks are assigned to different processors or cores, and each
processor handles its task independently, possibly with varying execution times.
 Example:
o In a web application, one task could be managing user input while another
task processes data, and yet another communicates with the database.
  Advantages: TLP allows for high concurrency and is well-suited for multi-core
processors. Unlike ILP, which focuses on parallelizing instructions, TLP focuses on
parallelizing entire tasks or threads.
 Applications: Useful in systems where different program parts can be divided into
independent tasks, like operating systems, web servers, and real-time systems.

4. Transaction Level Parallelism

 What It Is: This parallelism type is specifically used in database systems or
transactional environments where multiple transactions (distinct sets of operations)
can run simultaneously.
 How It Works: Different transactions are processed in parallel, often without
interference, as long as they don’t conflict over data.
 Example:
o In a bank database, one transaction might be updating a user’s balance, while
another transaction checks the account status of a different user. These can
happen simultaneously, as they don’t interfere with each other.
 Concurrency Control: TLP often requires mechanisms for concurrency control to
handle conflicts, such as two transactions trying to modify the same data
simultaneously.
 Applications: Primarily used in databases, financial systems, and any application
where multiple independent transactions need to run concurrently.

Flynn’s Classification of Parallelism

Flynn’s Classification is a way to categorize computer systems based on how they handle
instructions (commands) and data. It helps us understand the different ways computers can
work on multiple tasks or sets of data simultaneously. There are four types in this
classification: SISD, SIMD, MISD, and MIMD.
1. Single Instruction, Single Data (SISD)
 What It Is: The computer can process only one instruction on one set of data at a
time.
 How It Works: Every step is performed one after another, like following a simple
recipe with no shortcuts.
 Example: Traditional, single-core computers that perform one task at a time without
parallelism.
This type of system is straightforward but slower for tasks that could benefit from parallel

processing.

2. Single Instruction, Multiple Data (SIMD)

 What It Is: The computer performs one instruction but applies it to multiple sets of
data at the same time.
 How It Works: Think of it like a teacher (instruction) giving the same direction to a
group of students (data points) all at once, with each student performing the action
individually.
 Example: Graphics Processing Units (GPUs) use SIMD to process many pixels in
parallel, making it great for tasks like image and video processing where the same
operation needs to be repeated across lots of data.
SIMD is efficient when you need to perform the same operation on a large amount of similar
data.
3. Multiple Instruction, Single Data (MISD)
 What It Is: The computer can perform multiple instructions on the same set of data
at the same time.
 How It Works: Imagine several doctors (instructions) looking at the same patient
(data) to give different treatments or evaluations.
 Example: This setup is rare in practice but can be used in fault-tolerant systems,
where multiple processors analyze the same data to detect errors.
MISD is uncommon but can be helpful in specialized systems that need to ensure data
accuracy or reliability.

4. Multiple Instruction, Multiple Data (MIMD)

 What It Is: The computer can perform different instructions on different sets of data
at the same time.
 How It Works: This is like a team of chefs in a restaurant, each preparing different
dishes using different ingredients at the same time.
  Example: Most modern multi-core computers are MIMD, where each core can work
on its own task with its own data, making it ideal for running multiple programs or
complex tasks that can be split into smaller parts.

MIMD systems are very powerful and flexible, allowing for true parallelism by handling
various tasks at once across multiple processors.

ARM PROCESSOR

The ARM (Advanced RISC Machine) processor is a type of CPU known for its energy
efficiency, simplicity, and versatility. ARM processors are widely used in mobile devices,
embedded systems, and increasingly in servers and laptops. Here are the main features of
ARM processors:
1. RISC Architecture (Reduced Instruction Set Computer)
 ARM processors use a simplified set of instructions compared to CISC (Complex
Instruction Set Computer) architectures, such as x86.
 The reduced instruction set leads to faster processing and lower power
consumption.
 This simplicity makes ARM processors ideal for battery-powered devices.
2. Energy Efficiency
 ARM CPUs are designed to be power-efficient, which is why they’re widely used in
smartphones, tablets, and IoT devices.
 Their low power consumption allows for longer battery life, making them suitable
for mobile and embedded applications.
3. 32-bit and 64-bit Variants
  ARM processors come in both 32-bit and 64-bit versions, allowing for flexibility
depending on the application needs.
 The 64-bit ARM processors can handle larger data sizes and address more memory,
which is essential for modern computing requirements.
4. Multiple Cores and High Scalability
 ARM processors are available in single-core to multi-core configurations, allowing
for a range of performance needs.
 They can scale from simple devices like microcontrollers to powerful, multi-core
processors for servers and desktops.
5. Thumb and Thumb-2 Instruction Set
 The ARM processor includes a Thumb instruction set, which provides 16-bit
encoding for frequently used instructions, reducing memory usage.
 Thumb-2 combines 16-bit and 32-bit instructions, allowing for greater code density
and efficiency, especially in memory-constrained environments.
6. Floating Point and SIMD Support
 ARM processors often have a Floating Point Unit (FPU) for mathematical operations,
beneficial for media and scientific applications.
 Some ARM CPUs also support SIMD (Single Instruction, Multiple Data) operations,
enhancing performance for data-intensive tasks like graphics processing.
7. ARMv8-A and ARMv9 Architecture Enhancements
 ARMv8-A introduced 64-bit processing, improved cryptographic extensions, and
enhanced virtualization support.
 ARMv9 builds on ARMv8 with enhanced performance, improved security
(Confidential Compute Architecture), and enhanced machine learning capabilities.
8. Security Features (TrustZone)
 ARM TrustZone technology provides a secure environment within the processor to
run trusted code, separating it from the main operating system.
 This secure environment is essential for secure transactions, DRM, and other privacy-
sensitive applications.
9. Vector Processing (NEON Technology)
 Many ARM processors feature NEON, a technology designed for accelerated
multimedia and signal processing tasks.
 NEON supports parallel processing of data, ideal for image processing, video
encoding, and audio applications.
10. Virtualization Support
 ARM processors, especially in the ARMv8 and newer architectures, include
virtualization support.
 This allows for running multiple operating systems on a single processor, useful in
server environments and development.