KATHOLIEKE

UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE

Multitasking Design and

Implementation Issues in
Embedded Systems
ESC Europe 1999 report

by Larry Mittag

1
 Overview
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE
 Speaker
 Larry Mittag (Stellcom Technologies)
 Presentation
 first part was basic but interesting
 second part: hints & tricks
ESC Europe 1999 report

 no groundbreaking concepts

2
 Part 1: Properties of RTOS’s
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF

 Linear code vs. multitasking

COMPUTER
SCIENCE

 Multitasking and multithreading

 Use of multitasking and multithreading in embedded
systems
ESC Europe 1999 report

 Communication
 Priorities

3
 In The Beginning: Linear Code
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE
 Simple and effective
 Augmented by interrupts
 Preferred method for systems dedicated to simple
tasks
ESC Europe 1999 report

 But this can lead to an overloaded single task

4
 Next Step: Event Loops
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE

while (1)
{
ESC Europe 1999 report

Do_Task_1();
Do_Task_2();
Do_Task_3();
}

5
 Next Step: Event Loops
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE
 Allows to handle multiple logical tasks in one thread
 Nice design to keep separate code segments away
from each other
 Can eventually fall victim to complexity
ESC Europe 1999 report

 ad hoc techniques lead to this complexity

6
 Asynchronous Multitasking
KATHOLIEKE
UNIVERSITEIT
LEUVEN

 Provides a finer degree of sharing of the CPU

DEPT. OF
COMPUTER
SCIENCE

 Relieves the programmer of the need to `order`

different tasks
 Complex in sense of
 design
ESC Europe 1999 report

 debugging

7
 Multitasking Flavors
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE
 Lightweight multitasking
 another word for multithreading
 Heavy-duty multitasking
ESC Europe 1999 report

8
 Lightweight Tasking: Threads
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE
 Address and global name space shared among tasks
 Separate stacks, so local variables are not shared
 Very fast context switches
 only need to save registers and stack pointer
 No memory management required
ESC Europe 1999 report

 no need to dereference pointers

9
 Monolithic Load Module
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE
 All (or most) references resolved at link time
 Difficult to add load modules at runtime
 Very simple and fast
ESC Europe 1999 report

10
 Heavy-Duty Multitasking
KATHOLIEKE
UNIVERSITEIT
LEUVEN

 Each task has a separate virtual address space

DEPT. OF
COMPUTER
SCIENCE

 Also a separate global name space

 Generally requires memory management unit on
CPU
ESC Europe 1999 report

11
 Dynamic Load Capability
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER  Allows new tasks to be loaded from disk or
SCIENCE
communications link
 Requires virtual memory, position-independent code,
or trickery
 Not necessary in many systems, required in others
ESC Europe 1999 report

12
 Pros and Cons of Multithreading
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE
 Pro
 Simple, fast multitasking
 Easy, fast communications through global variables
 Contra
 difficult to implement dynamic loading
ESC Europe 1999 report

 a bug can be disastrous

13
 Pros and Cons of Multitasking
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE
 Pro
 Very easy to build smaller, more easily debugged tasks
 Damage from a misbehaving task can be contained
 Contra
 overhead at runtime
ESC Europe 1999 report

 complex CPU
 may require swap medium

14
 Which for Embedded Systems?
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE
 In general, go with lightweight threads
 Exceptions:
 Virtual memory required
 Need to use large amounts of ported “workstation” code
 Need to dynamically load external code
ESC Europe 1999 report

15
 When NOT to Use Multitasking
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE  Small, simple applications
 Applications that need to be validated
 When determinism is important
ESC Europe 1999 report

16
 Communications
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF

 Shared memory
COMPUTER
SCIENCE

 fast
 but: deadly
 Semaphores
 simple
ESC Europe 1999 report

 but: deadlocks
 Message queues
 decoupling sink and source
 but: out of sync consume/produce rates

17
 Task Prioritization: Crutch?
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF

 Prioritization allows the specification of the relative

COMPUTER
SCIENCE

importance of tasks
 Can be very important - allows the use of
“background” processing
 Also can act as a crutch for a poor system design
ESC Europe 1999 report

 last moment solutions

 politics

18
 Why Prioritize?
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER  Creates hierarchies of task importance
SCIENCE

 Gives critical functions a better chance of being

deterministic
 One more way the programmer can exert control
over the system
ESC Europe 1999 report

19
 Priority Overkill
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF

 More than 3 or 4 levels are rarely useful

COMPUTER
SCIENCE

 Overuse can cause inadequate levels of task

switching
 Potential result: Poor response time
ESC Europe 1999 report

20
 Rules for Sane Prioritization
KATHOLIEKE
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE  Divide the tasks into time-critical and non time
critical
 Among critical tasks, give short-duration tasks
higher priority
ESC Europe 1999 report

 Allow your system to be a little out-of-control

21
 Part II: Designing Multitasking
KATHOLIEKE
Systems
UNIVERSITEIT
LEUVEN
DEPT. OF
COMPUTER
SCIENCE

 Selecting an RTOS
 Partitioning tasks
 Communications and synchronization
ESC Europe 1999 report

 Adjusting tasking parameters

 Debugging techniques for tasks
 Grab-Bag of stuff

22