Message routing in multi-segment FTT networks: the isochronous approach Paulo Pedreiras, Luís Almeida {pedreiras,lda}@det.ua.pt Workshop on Parallel and

Message routing in multi-segment FTT networks:

the isochronous approach

Paulo Pedreiras, Luís Almeida

{pedreiras,lda}@det.ua.pt

Workshop on Parallel and Distributed Real-Time Systems 2004 (WPDRTS04).April 26 th and 27 th , 2004, Santa Fe, New Mexico

DET – IEETAUniversidade de AveiroAveiro-Portugal

26-27 April, 2004WPDRTS 2004, Santa Fe, New Mexico2

Message routing in multi-segment FTT networks: the isochronous approachPaulo Pedreiras, Luís Almeida, WPDRTS’04

General framework

Industrial systems (in a broad sense) are more and more integrated

As the system size grows, so does its complexity A possible approach to handle complexity is to build

the system by composing subsystems

Breaking a large network in segments may:– Facilitate the system management– Increase the traffic schedulability level– Isolate independent traffic– Allow the physical extension of the network



General framework - 2

Communication across different subsystems takes place through gateways



General framework - 3

For real-time applications it is necessary to guarantee the schedulability of both:

– Intra-network traffic– Inter-network traffic

Category of problems addressed in several contexts:– Voice and video on WANs– Multi-computer systems interconnected by mesh networks– Wireless networks– Switched Ethernet networks– ...



Contribution

Multi-segment support to the Flexible Time-Triggered communication paradigm (FTT):

– Comparison, in terms of end-to-end latency, between Isochronous and Anisochronous architectures

– For the isochronous architecture: Two deadline allocation strategies:

– Isometric– Maximum Schedulability Laxity

and comparison of their relative performance



FTT brief overview

The FTT paradigm main operational characteristics:– Centralized scheduling with operational flexibility– Master/Multi-slave cooperation model– Support for distinct traffic classes

Event /Time-Triggered traffic, with temporal isolation Hard/Soft/Non real-time timeliness requirements

How it works?– Traffic is allocated in fixed duration time slots

( Elementary Cycle - EC ) – Bus time is organized in an infinite succession of ECs– ECs start with a trigger message (TM) sent by the Master



FTT brief overview

Elementary Cycle structure

Synchronous window Conveys the time-triggered traffic The TM contains the EC-Schedule

Asynchronous window Event triggered traffic,

real and non-real-time



Isochronous vs anisochronous architectures

Non synchronized FTT segments may lead to high end-to-end latency (synchronous traffic)



Isochronous vs anisochronous architectures Synchronized FTT segments may lead to lower end-to-end latency

End-to-end deadline equal to sum of intermediate deadlines



Isochronous vs anisochronous architectures

Isochronous architecture

– Requires clock synchronization

CPU overhead Communication overhead

– EC lengths constrained to be harmonic

– Tight control on the inter-network traffic latency

– Reduced end-to-end latency

Anisochronous architecture

– Lower CPU/Network overhead

– Unconstrained EC length

– Lower efficiency in inter-network traffic handling, leading to a higher end-to-end latency



Deadline allocation scheme

The problemGiven a message end-to-end deadline, how to compute the

intermediate deadlines in each one of the involved networks?

System model– FTT isochronous networks

– Interconnection via gateway nodes that fully comply with the

FTT trasmission control policy

– Synchronous message i of network j characterized by: SMi,j={Ci,jPi,j Di,j Pri,j i,j}

– Die2e : end–to-end deadline of message i




For each message i having to cross networks 1..k:– Latency:

Goal:

and feasible message sets in each one of the intermediate networks

K

j jiDi

L 1 ,

eeii DL 2




Isometric allocation scheme– Message deadline equally divided between all the involved

networks

– Simple computation– No need to know global system state

but …

– May lead to bottlenecks

K

DDf

eei

jiKj

2

,..1 ,




Maximum schedulability laxity– Assign deadlines according to the relative network workload

– Normalized utilization:

iav

ii U

UU '




Maximum schedulability laxity (cont)– Deadline computation:

– Compared with the isometric strategy: Requires global data (individual network utilization) More complex ( O(k) instead of O(1) )

but …

Higher schedulability Best suited for systems requiring on-line QoS management

– Load balancing

eeiK

nn

jji D

U

UDf 2

1

'

'

, *



Simulation results

FTT implementation on CAN– EC =10ms– Bit rate=125kbps– FTT overheads = 7% / EC– Messages between 1 and 8 data bytes– Periods between 10 and 60 ECs, Deadlines=Periods– Number of networks between 1 and 5

Asynch. window length(% of EC)

Initial network util.(%)

Schedulab. level(% of synch. bandw.)

min max min max min max

5 20 15 40 50 90



Simulation results

Number of scheduled messages

Average network utilization ratio



Conclusion

There may be advantages from using segmented time-triggered networks

Reduncing latency of inter-segment traffic requires global synchronization

– Isochronous vs Anisochronous architectures– The Isochronous architecture provides a better control of

inter-network traffic latency

Two methods to compute inter-network message deadlines:

– A simple isometric allocation scheme– An allocation scheme that partitions the deadline according

to the leeway of each intermediate network

Documents

Message routing in multi-segment FTT networks: the isochronous approach Paulo Pedreiras, Luís Almeida {pedreiras,lda}@det.ua.pt Workshop on Parallel and