Real Time Scheduling Issues in Powertrain Controls James B. Kolhoff Engineering Group Manager Front

1
Real Time Scheduling Issues in Powertrain
ControlsJames B. KolhoffEngineering Group
ManagerFront Wheel Drive Controller TeamGeneral
Motors Powertrain
2
Overview of Presentation

• GMPT Electronics Integration SW - Group
  Product
• Scheduling Requirements and Problem
• Solution
• Distributed architecture
• Next step

3
Group and Product Background

• Group
• GMPT is a division of General Motors, responsible
  for engine, transmission, powertrain controls
  engineering and manufacture
• Electronics Integration Software (EIS) is a
  product engineering team responsible for the
  electronics and software for powertrain controls
• Product
• EIS end product is an embedded microprocessor
  control module(s) that controls and diagnoses
  engine, transmission, and vehicle functions.
• Multiple end products (ECM, TCM, PCM) with
  different feature content (internal GM and
  external customers)
• Multiple controller and compiler suppliers
• Other vehicle module interfaces
• Development and production tool interfaces
• Controller 32bit uc, 1Mb ROM, 150 pins

4
Scheduling requirements

• Two categories of task - time synchronous, engine
  event synchronous
• Time 3.125ms, 6.25, 12.5, 25, 100 ms
• Engine crankshaft synchronous, cam synchronous
• The engine event tasks cause the processing power
  to be consumed in direct proportion to engine
  speed
• Engine event synchronous tasks have harder
  deadlines and higher priorities than time based
  tasks
• 8 cylinder engine engine, event sw task execution
  time 1ms
• 600 rpm 25ms event rate, 4 available
  processor thruput
• 7000 rpm 2.1ms event rate, 48 available
  processor thruput

5
Task scheduling
6
Scheduling Problems

• Most critical scheduling problem was task
  deadlines missed at higher engine speeds
• Basic root cause Limited processing power
• Using low cost microprocessor
• Low clock speed for EMC performance
• Too late in program to make processor change
• ROM limited so we cant do ROM tradeoffs for
  thruput
• Fixed point math operations
• Library not optimized for performance
• Requirement of ANSI-C for code portability
• Not designed for performance
• SW Design and Coding Standards
• Designed for reuse and readability, not
  performance

7
Solutions applied

• Re-design software for improved efficiency
• Significant work effort, potential loss of
  function, repeat verification
• Optimize libraries to take advantage of processor
  specifics
• Significant work effort, reduces reuse, increases
  verification requirements
• Revise coding standards to maximize efficiency
• At the expense of portability and reuse
• Rework and revalidation across large number of
  engineers
• Biggest bang for the buck - dynamic scheduling
• Can localize redesign at areas of maximum benefit
• Time tasks slower than 25ms rates are
  insignificant to the problem

8
Dynamic scheduling

• Objective Reduce execution requirements at
  higher engine speeds
• Difficult to individually disable or redesign
  functions
• Developed engine speed zones approach
• Different function level in each zone
• Simplifies coordination of scheduling change
• In middle engine speed range, divide function
  across multiple engine events
• Balance load across multiple cylinder events
• At highest engine speeds, significantly simplify
  some functions
• Engine states dont change every cylinder

9
Effect of Dynamic Scheduling
10
Controller System Topologies
Powertrain Control Module
Engine/Transmission Control Modules

Engine Electrical System
Engine Electrical System
ECM
Vehicle Electrical System
Vehicle Electrical System
PCM
Transmission Electrical System
Transmission Electrical System
TCM
11
Distributed architecture

• Controller systems architecture for GMPT is
  changing to separate engine controller /
  transmission controller
• For reasons of powertrain portfolio management
• This architecture reduces the computing power
  needed in any single controller
• Scheduling and thruput still needed to be
  carefully managed
• System partitioning plays a key role
• Inter-module Communications uses some of the
  freed up thruput

12
Future

• Microprocessor power has grown dramatically over
  the past 5 years
• At the same time, costs have fallen dramatically
  for this power
• With the microprocessors available for the
  projects planned, thruput will not be the
  significant problem it has been in the past
• Simulation and schedule/thruput budgets are the
  next steps