A Tool for Describing and Evaluating Hierarchical Real-Time Bus Scheduling Policies

Author: Trevor Meyerowitz, Claudio Pinello, Alberto DAC2003, June 24,2003

Presenter : Tsung-Yu Ho(Seminar Book Page 81)

Abstract

We present a tool suite for building, simulating, and analyzing the results of hierarchical descriptions of the scheduling policy for modules sharing a bus in real-time applications. These schedules can be based on a variety of factors including characteristics of messages and time slicing and are represented in a hierarchical tree-like structure that specifies multiple levels of arbitration. This structure can describe many popular arbitration schemes. Our simulator evaluates the specified scheduling structure on a set of message traces for a given bus.

We illustrate our approach by applying it to two examples: the SAE Automotive Benchmark and Voice Over IP (VoIP). Although this paper deals with just bus scheduling policies, the approach can be easily extended to other real-time scheduling problems.

Outline What’s the Problem Introduction

Scheduling policies Propose a tool suite

Related work Various scheduling

Scheduler representation and evaluation Define representation Describe in language Evaluation

Experiment Result SAE Benchmark VoIP Benchmark

What’s the problem

Increased complexity of interaction between blocks

Previous work focus on selecting process mapping and communication topology Ignore arbitration policy of bus.

In real-time system, most applications have real-time constrained, it is hard to analyze.

Introduction

Focus on the representation and evaluation of various scheduling policies Real-time message among modules communicating

via a shared bus.

Propose a tool suite (called STRANG) Building, simulating, and analyzing the result of

scheduling policy Provide a simple hierarchical language

Describe the arbitration policy Simulate the policy Easy to explore the design space

Scheduling Policies (Related work) This paper is based on well-known scheduling Focus on communicating scheduling (via shared bus) Introduce the popular scheduling as follows:

1.Event-Triggered Scheduling

2.Time-Triggered Scheduling

3.Hybrid Scheduling

1. Event Triggered Scheduling

Based on priority FIFO ordering : simple to implement Fixed Priority : simple to implement EDA (Earliest Deadline First) : best result

Dynamically give priority to message

CAN (Control Area Network) bus Successfully used in event triggered scheduling Use a fixed priority arbitration scheme

Based on id message (id : identifier) Each node can use bus when there is no message

being transmitted. Has more flexible than Time-triggered scheduling

2. Time-Trigger Scheduling

TDMA (Time Division Multiple Access) Policy Divide a period time into several time slices Each time slice is assigned to only one node

Easy to ensure fairness between the nodes

TTP (Time-Triggered Protocol) bus Use the TDMA policy

A lower arbitration overhead than CAN bus Higher bandwidth utilization

Easy to have latencies Because of non periodic message

Inflexible than CAN bus

3. Hybrid Scheduling

Improve Performance by combining Event-Triggered and Time-Triggered Scheduling

Provide the flexibility of CAN with determinism of TTP

Hybrid approaches usually is used in Multimedia domain (like VoIP)

Outline What’s the Problem Introduction

Scheduling policies Propose a tool suite

Related work Various scheduling

Scheduler representation and evaluation Define representation Describe in language Evaluation

Experiment Result SAE Benchmark VoIP Benchmark

Definition A hierarchical arbitration policy schedule

message between p entities communicating via a shared bus P : primary node M : Message ZB: Scheduling Policy MB: a trace on the bus

Goal Pick the best policy

under fitness metric

Characteristics Message and Nodes

Each message has its own priority Sender-id, receiver-id, size, message-id, arrival-time

deadlines of message, time until deadline. Primary Nodes choose available message to bus Bus scheduling policy determine which message go

first Metrics

Find the quality metric The number of missed deadlines The overall execution time The average throughput of the bus

Then choose fitness policy

Tree Representation Arbitration Syntax

P : first custom operation A : arbitration node S : primary/sender node PolicyID: policy name used by node Alloc : the style of time allocation used by node Preemption : preemption policy

Sample Trees CAN Tree

Event-Triggered

TTP Tree Time-Triggered

Hybrid Tree Combined two of these

Examples of TTP

Experiment Result (SAE Benchmark) SAE (Society of Automotive Engineers)

Different protocol at bus speed 100Kbps, 125Kbps, and 250Kbps

53 message type between 7 node 5 second of message trace

SAE Benchmark Evaluation Result (1/1)

SAE Benchmark Evaluation Result(2/2)

Voice over IP Benchmark

G.729A voice codec 10 byte samples every 10 ms

Evaluate 4 types of arbitration policies EDF, FIFO Fixed Priority with RMS (Rate Monotonic

Scheduling) non-preemptive, shortest periods have the highest priorities.

Fixed Priority with DMS (Deadline Monotonic Scheduling) shortest deadlines get the highest priorities

Conclusion

Formulate the problem of scheduling real-time messages on a shared bus

Show the benefit of using hierarchical arbitration policies for optimizing the schedule

Present a tool that can represent a wide variety of trees, and simulate them using message traces.

The results could easily be generalized to evaluate scheduling policies in a variety of other domains.