(.4~p\lig lll
\'0
11-" .\(
C411111411. Cll l llPlllt'I",
( :41111l1ltll1i, ;lIioll .... ill
11 . 111"\HI1I ;lIi4 HI.
1'. t I i.... . " I: I! It l', I ~ I."~ I
TEST AND DEVELOPMENT BENCH FOR AN IN-BORNE COMPUTING ARCHITECTURE
J.
Bancelin and H . Bo rde nave
I(-\ FI'. I)
Abstract .
The
RATP
investigations,
(Rail)
research
1"//1 '
./lI/n \ ',,11,-. ,. ;)(1/ I 1',11;'. F I"II 11 11 '
Rolling Stock Department set in motion a whole range of
projects
and
experiments
in
1982
aimed
at defining a
computing architecture for future rolling stock . The present article describes the methods and means deployed for evaluating the computing architecture and in- borne local ne tworks for MF 88 and MP 89 rolling stock scheduled for introduction on the RATP Metro System as from 199 1. Keywords. Computing architecture, Reconfiguration , Local networks .
In
198 2 ,
the RATP (Rail) Rolling Stock Depa rt-
safe t y
informati on
(emergency
ment initiated a whole ranqe of investigation s , research projects and experiments coverinq the
closing vehicle
computi ng
distributed computing hardware.
architecture
of
rolling stock . This
operations, cahling
braking ,
door -
etc . . . ) and lim iti ng using series - connecteo
exe rcise is now entering its practical implemen-
High
tation phase, which coincides with th e recent o rder p lace d for MF 88 stock (destination: Line 7bis) and with tenders i nvi t ed for unmanned MP 89 stock ( des tination: Line 11 ). This paper problem :
will
examine
availability
levels
are
secured through
failure toleran ce . Hence the ongoing examinati on of automatic reconfigurations whereby an y function built into a failed computer is automa tically taken over by a sound computer even if performance levels are adversely affected as a
three aspects of the
result.
This
availability consequently implies
rlistribution of the computing system .
I", at lies behind the computing architecture option 7
High
maintainability
through
How
best
to
implement
and
validate this
levels should be achieved
for "first - level
ll
maintenance
architecture ?
requirements (replacement of a failed removable element) with set objectives as regards d iagno-
Development
test
sis
of a test bed for the computing
accuracy .
apparatus
This diagnosis is effected using
integrated
with
fun c tional
equipment.
architecture.
1.
cat e ring
Lastly, cost reductions are achievable thro uqh standardisation of computing hardware, in other words by reducing the number of cards, usinq
What lies behind the computing architecture option
instead cards and components available on the market duly adapted to all ow for rail - specifi c The latest genera tion o f rolling stock (MF 77 and MI 84) has become increasingly sophisti cated , difficult to develop while also o ff ering an unsatisfactory quality- of - service level.
cons traints.
The
four
ob jectives
require
development o f a
g l o bal vision o f the train, which must no lonqer he consinered as the juxtaposition of equipment ite ms studied in isolation and then assembled,
The study objectives are fourfold
but rather as a system nesigned to fulfil clearly- defined functions under optimum condi tions while ensurin g set availability l evels.
reliability improvement , availability levels c ons i s tent with int egra l
To t h is end, adopted :
automation c riteria,
the following criteria have been
maintainability improvement , functional breakdown of the train into main "themes ", by doing away with the traditional
investment and maintenance cost reductions .
physical breakdown , In
order
f or reliability to improve, two condi~ions must first be fu l filled: cabling must be simplified and the number of conne ct ing poi nts reduced. The solution l ies in replacing the different train lines by an in- borne local network
characterisation of the information exchanged between these different "themes" .
This approach fac i litates conception of a standard computer model (hardware and software)
- except for some twenty lines carrying
1.-,
.I-
16
Ibll(c lill alld 11. Bordt'Il;I\t'
f or any t rain function, the l oca l n etwork and the other links. The archi t ecture study consists then of defining the distribution of computers on the train and the distribution of functions within each computer by seeking optimisation
user applications,
regularity of message transfer nistic or probabilistic) ,
(determi-
serv i ce offered to users, of requirements, of reconfigurations, costs.
of the network load, The first two criteria (reliability, durability) lead to favouring the use of existing standards in respect of l ocal network s . This would suggest excluding specific networks wh i ch identify too closely with a manufacturer or equipment assem-
of maintenance .
2. How best to imp l ement and validate this
b l er ,
architecture
added
to
which these II factory " networks
genera lly imply using electronic manufactured by the same firm. In order to test the different computingarchitecture op tions between them, RATP decided to deve l op a range of simulation equipment designed to compare and test different computing architectures for rolling stock.
equipment
As part of this standardisation process , the functions performed by the communication system are consequent ly examined against the backqround
of the ISO's OS I reference mode l s, with these functions being broken down into 7 layer s each
The end purpose of these simulations is to obtain validation of a computing architecture in
performing a specific funct.ion .
order to achieve
In the context of these OSI l ayers , emphasis has been placed essentially on the first two (Physical layers an d Date link). In practice, these two layers are frequently obta ined using
optimisation sing
of the distribution of proces-
operat io ns
relating to the functional
themes
(driving ,
doo rs,
energy ,
safety , etc ... )
dif f erent physical the train) ,
motive assured
computers
power , by the
available on
op timisati on of the geograph ical distribution of these phys i cal computers) .
components by using
Hence
2.1 - Phase 1
Network evaluation mechanism
by
Computers,
virtue
of
their
distribution ,
intercotnrrnlnicate thr ough a "local " communication network .
This network forms the backbone of the train system. Architecture validation therefore starts with evaluations of the communications networks , followed by a choice. These evaluations address the external appearance of the netwo rk, in other
words
from
rea ction
the
with
standpoint respect
of
-
specialist
chips -
the
decision
to
limit
to subscriber behaviour
2. 1.1 - Approach to l oca l-network preselection
therefore
based
on
criteria
tions . Where
the
token rin g is concernen ,
bus) , no network was se lecte d , r easons explained below :
Selecting an industr ial l ocal network is a task that requires several criteria to be taken into consideration :
of communication - system pe rfor-
mance ,
investment (20 years in
this pa rti cu lar instance), sturdiness to cope with an aqqressive environment (ele ctromagnetic interferences, mechanical and temperature constraints) , . speed
of
information
transfer between two
the network
selected is the TORN AD (manufactured by ALSTHOM ) model adopted for the Atlantic TGV. This network, although not strictly in conformity with standard 802.5, was preferred because of its rail techn ology . Lastly ,
the
are more
\,e have selected , in standard 802 . 3 , networks ARLIC (developed by SEMAGROUP , ethernet), FACTOR (manufactured by APTOR , determinist ic ethernet) which offer guarantees of the t ype obtained with tried- and- tested existinq industrial installa-
l ocal networks which could potent ially be train installed, with the defi nitive cho ice among preselected networks being effected dur in g the next phase .
of
that
strategic than strict l y technical .
The aim of the exercise is to pre - select various
durability
examination to
networks whose base layers ( 1 and 2) arguably correspond to international standards IEEE 802 . 3 (C SMA) , IEEE 802 . 4 (token bus), and IEEE 802 . 5 (token ring). This initia l s orting process is
users (network
patterns) .
reliabili ty
much more than
software . However , the problem of hardware durabi lity is much more acute than that of software durability (comp onent obsolescence , higher cost of rep l acement due to the multiplier effect of lar ge series).
with
regard
to
standard 802.4 (token for
severa l
The IEEE 802 . 4 context is essentially linked to the MAP network , the general concept o f which (factory-oriented) is of little relevance
to
a train environment where the
MAP higher-order services ( general level, control r oom) hardly exist or do no exist at all . OUr needs would seem to identify more close l y with concepts o f the Mini - Map , Proway type , etc ••• ). (NOTE : this MAP general concept cou ld nevertheless be va lid for a Metro line. It does not ne cess arily imply using a token bus) • MAP stabilisation in version 3 . 0 is a recent
development, and doub ts exist over the l o n gterm du rabulity of an investment undertaken
IIl-horlll' ( :O IIlPllli l l~ .\rcil i ll'ctllrt'
today.
in stage one: the transfer functions of each functional theme are simulated by means of software functions. It then becomes pcssible to simulate and evaluate several computing architectures for the particular process control being studied .
All industrial references as regards MAP networks are still relatively sparse in terms of tried- and-tested sites. Last l y, various studies undertaken (by INRIA in particular) would suggest ex i stence of a weakness
in
"token"
networks
This phase ends with a choice of architecture (phase 2).
placed in a
highly-perturbed environment. While in no way wishing to get involved in technical arguments here , this finding does add weight to our contention that the validity of such networks must be demonstrated on the basis
To conclude on this point, it can be said that the MAP conceptual approach (and not standard 802.4) could prove highly pertinent in the coming years but is today somewhat premature given
our
specific
in stage two the selected architecture is developed and validated (phase 3).
2.3 - Phase 4
of existing sites.
requirements
in
terms of
durability more particularly.
2.1 . 2 - Final selection
17
A
subsequent
Reference chain
phase
consists
of
replacing
transfer-function simulations by real hardware. We then obtain a reference chain that makes it
possible to develop the actual computing architecture prior to placing of the rolling stock in service , through gradual integration of effective hardware. During the service life of this rolling stock , this same reference chain is used
to study and validate modifications. This phase involves selecting, among the three networks mentioned above (FACTOR , ARLIC, TORNAD) the model that will e ffectively be train instal led. In practive, this selection will be based on the network evaluation process described below (paragraph 3).
3 - Test bed for computing architecture
RATP has commissioned from MATRA TRANSPORT a test bed codenamed TERASA which is designed to facilitate simulation and validation of the inborne computing system throughout its main design phases as defined earlier ,
namely
The selection criteria are those already descri -
bed earlier : Speed
user
of
information
applications.
transfer between two
Application
to
Appli -
Phase Phase 2 Phase 3 Phase 4
Testing and selection of local network Computing architecture selection
Architecture validation Reference ch ain
cation message transfer time.
Reliability viour
of
communication - system
(percentage
of
lost
or
beha -
erroneous
messages) • Sturdiness ronment :
The architecture of this test bed depends necessarily on t hat of the system to be simulated . ~.Je are dea lin g here with a case of
distributed computing . to cope with an aggressive envi-
TERESA consequently ration (fig .1 ) :
has the following configu-
- Electromagnetic interferences, - Sectioning of medium/media, - Failure of connector cards.
train computing resources (actual in - borne equipment) are simulated by some 15 "real time" computers u sing MC 68020 microprocessors. Each computer simulates one or several
Message transfer regularity
functions to be performed aboard the train,
This
is the guarantee offered by the commu -
nication
system
over
delivering the
information within a set time bracket. The approach at the level of low layers can be deterministic (APTOR, TORNAD) or probabilistic (ARLIC). At application level , determi nistic greater account
regularity seems to carry much implementation uncertainty on of the combinational nature of
transfer cases and times through material and software layers. This generally leads to quasi- probabilistic operation which can be quantified thanks to the platform.
2.3 - Phases 2 and 3
Architecture evaluation and selection mechanisms
Once the network selection has been carried out, the architecture validation phase proper is set in motion in two distinct stages :
each
one
of these comput ers is linked into
the local network effectively used phase 1,
this
of the train, which is and not simulated . In
l ocal network can be anyone
among the networks ARLIC , TORNAD) ,
to
be tested (FACTOR,
the whole set of 15 computers is linked to a SUN/UNIX - type work station via an Ethernet/ TCP - IP service network. This station then acts as IIban d-leader " for the whole simulation exercise, and repatriates the measurements carried o ut by each II rea l-time" computer .
Analysis and recording of measurements taken are performed at this work station . Selective search for a particular test or group of tests is performed with a SGBD (Oracle). In practice, Phases 1 to 4 described above are implemented as follows :
.1.
IH
Ildll( "Iill
;tile!
11. H()nkll;II('
Phase 1 (test network) : the operator specifies the traffic to be generated on the network (volume of date for transmission, message type, transmission moment) for each transmitterreceiver pair involved in the simulation. Performance measurements are obtained by notinq the moment of issue of messages from their
transmitter
application,
and
that
of
their
arrival to the receiver application. The communication system transfer time is then deduced.
Phases 2 and 3
(architecture
test) : the func-
tionalities to be assured by software tasks are simulated. Moreover, the physical environment of
the
system
is
also
the object of a computing
simulation (mechanical inertia of train during a
speed measurement phase). Phase 4: equipment
in this phase, the actual train systems are mounted. They dialogue
with environment sirnulations performed during phase 3, via an interface electronic device so as to recreate a complete looped system representing the equipment of a train in service.
Station
XMS 30tO VME
Int eractive
Temps reel
Reseau de pilotage ...------------ Et h e rn e t