Real Time Systems and Soft Computing

We investigate the potential of using “soft computing” properties of kernels in this report for real time systems. The main attribute of “soft computations” is the fact that they can offer cruder results before they complete, or they might execute for a long period refining a previously adequate result. This thesis deals with the design challenges associated with creating a system which makes use of the “soft computing” qualities of kernels to boost real time performance. The classic technique to the design of real time systems is to figure out worst-case scenarios for the system statically and manually after which build the system with plenty of resources to fulfill deadlines and objectives. This technique was successful for traditional real time systems which function in uncomplicated, well-characterized environments. Having said that the emerging generation of complex, dynamic and uncertain real-time application domains stresses the growing requirement for flexible, adaptable design methods for real time systems. With the growing sophistication of real time systems, it is becoming infeasible to develop systems with adequate resources to accomplish the functional and timing demands of all application tasks consistently. Exactly what is becoming more and more significant in the new paradigm of real time computing is the demand to match deadlines with sufficient system solution quality without needing to develop the system to support worst case program execution.

Watch a video on Soft Computing


1. Introduction and Motivation
1.1 Real time systems: The current paradigm
1.2 Real time computing: The next generation
1.3 Reference Scenario: Sensor fusion for situation assessment
1.4 Motivation: Soft computations and performance accuracy trade-offs
1.5 Thesis contributions and organization
2. Background and Terminology: Real time Systems
2.1 Real time computing

Video on Concepts of Real Time Systems

2.2 Deadlines and Periods
2.3 Handling Aperiodic tasks
2.4 System Models
2.5 Hard and soft real time systems
2.6 Types of real time tasks
2.7 Predictability and Determinism
2.8 Real time scheduling
3. Related Work
4. Detailed Problem Formulation
4.1 Reference real time scenario: Northrop Grumman’s challenge problem for sensor fusion and situational assessment
4.2 System prototype
4.1.1 Information Form Data Association Algorithm (IDA)
4.2.2 Northrop Grumman’s cognitive test bed
4.2.3 Application of IDA to the Northrop Grumman test bed
4.3 Real time problem formulation
5. Soft computations and their characterization
5.1 Properties of soft computations
5.2 Performance profiles
5.2.1 Quality metrics for characterizing soft computations
5.2.2 Building performance profiles for individual soft kernels
5.2.3 Representation of performance profiles of individual soft kernels
5.3 Identification and characterization of individual soft kernels in the reference system prototype
5.3.1 Observation update sub-kernel /task
5.3.2 Inference sub-kernel /task
5.4 Composing systems using performance profiles
5.4.1 Deriving composite performance profiles for soft kernels
5.4.2 Model for optimal composition of “soft computing” kernels
6. Scheduling Policies
6.1 Static soft policy
6.2 Dynamic soft policy
6.3 Conventional Dynamic Earliest Deadline First (EDF) policy
7. Experimental Results
7.1 Evaluation metrics
7.2 Experimental set up
7.2.1 Data sets used
7.2.2 Simulation of real time sensor report delivery to the application
7.2.3 Counting missed deadlines and measuring CPU idle time
7.2.4 Measuring accuracy
7.3 Experimental results
7.3.1 Data set naming convention
7.3.2 Variation in tracking accuracy with different scheduling policies
7.3.3 Variation in CPU utilization with different scheduling policies
7.3.4 Variation in missed deadlines with different scheduling policies
7.4 Sensitivity study for performance profiles
7.4.1 Use of dynamic performance profiles across time steps for a data set
7.4.2 Sensitivity to different number of entities
7.4.3 Sensitivity to entity behaviors
8. Conclusion and Future Work

Source: University of Maryland

Leave a Comment