|
Course Highlights
This 2-day course introduces tools and techniques for distributing code and writing parallel algorithms in MATLAB. The course shows how to use Distributed Computing Toolbox to increase both the speed and the scale of existing code. Attendees who are working with long-running simulations, or large data sets, will benefit from the hands-on demonstrations and exercises in the course. Topics include:
- Working with a MATLAB pool
- Speeding up computations
- Task-parallel programming.
- Working with large data sets
- Data-parallel programming
- Increasing scale with multiple systems
- Installation and troubleshooting (optional)
- Schedulers (optional)
Course Objective
The aim of the course is to provide an introduction to Parallel Computing Toolbox for distributing code in parallel for processing large data sets and reducing simulation time.
Who Must Attend
Engineers who wish to use multiple system to shorten up the simulation time and to be able to process large data sets
Course Benefits
Upon the completion of the course, the participants will be able to write parallel algorithm in Matlab
Prerequisites
Participants who have attended our 'Comprehensive MATLAB', or equivalent experience using MATLAB
Course Outline
Day 1
Working with a MATLAB Pool
Objective: This section introduces a parallel approach to running MATLAB code through the use of multiple MATLAB sessions. Interactive techniques for prototyping in a parallel environment are highlighted. The concepts in this section also introduce several ideas explored throughout the course.
- Evaluating performance
- Distributing code
- Additional MATLAB sessions
- Parallel for-loops
- Evaluating speedup
- Hardware utilization
- Running in batch
Speeding up Computations
Objective: This section outlines the key steps for running parallel computations in a batch environment. The emphasis is on interacting with the various Parallel Computing Toolbox objects to create and run jobs that run in batch.
- Terminology
- Schedulers
- User configurations
- Object hierarchy
- Creating jobs
- Evaluating performance
Task-Parallel Programming
Objective: This section identifies important considerations for programming task-parallel jobs including decomposing a problem and partitioning input. Through use of a hands-on example, it also explores various techniques typically employed to achieve speedup.
- Decomposing parallel problems
- Reading from files in parallel
- Partitioning input
- Aggregating function calls
- Memory-mapped files
- Considerations for parfor-loops
- Task-parallel and data-parallel applications
Day 2
Working with Large Data Sets
Objective: This section explores working with arrays in a parallel environment, with an emphasis on parallel algorithms. Spitting large datasets across multiple instances of MATLAB, as well as simultaneously performing the same operation on the various portions, will be key themes. This chapter concludes by running prototyped code in a batch parallel job.
- Parallel terminology
- Types of arrays
- Composite arrays
-
Distributed arrays
- Creating distributed arrays
- Using distributed arrays
- Indexing into distributed arrays
- Running in batch
Data-Parallel Programming
Objective: This section explores the important programming considerations for parallel jobs. In addition, this section introduces using the communication features in parallel jobs for creating special architectures to solve specific types of parallel problems.
- Message Passing Interface (MPI)
- Sending/receiving data
- Collective communication
- Global operations
- Parallel models
- Deadlocks
- Synchronization
- Parallel topology
- Systolic architectures
Increasing Scale with Multiple Systems
Objectives: This section demonstrates tools for harnessing the power of multiple systems on a network for running code. Highlighted in the chapter are techniques for working with a heterogeneous mix of systems, as well as special features available in a full cluster of system.
- System components
- Accessing other schedulers
- Schedulers
- Dynamic licensing
- File and path dependencies
- Job and task states
- Callback functions
- Dynamically distributed applications
- Development/debug workflow
- Performance concerns
Installation and Troubleshooting (Optional)
Objectives: This section is targeted towards system administrators and those responsible for installing and maintaining a cluster. Provided in this section is a walk through of product setup, as well as testing and troubleshooting techniques.
- Installation considerations and options
- Installing/starting the MDCE service
- Starting the Job Manager
- Starting workers
- System shut down
- Additional setup options
- MDCE defaults
- User configurations
- Testing and troubleshooting
Schedulers (Optional)
Objective: This section covers the use of various scheduler products as part of a distributed computing environment. Scheduler benefits and tradeoffs, in addition to steps for integration, are key components.
- Scheduler overview
- Third-party schedulers
- User configurations
- The Submit function
- The Decode function
- Submitting jobs
|
