Design of the Control System of TRISTAN
https://accelconf.web.cern.ch/accelconf/p81/PDF/PAC1981_2359.PDF
IEEE Transactions on Nuclear Science, Vol. NS-23, No. 3, June 1981
Design of the Control System of TRISTAN
H.Ikeda, K.Ishi, T.Katoh, T.Kamei, Y.Kimura, S.Kurokawa,
S.Shibata, M.Takasaki, S.Takeda and K.Uchino
KEK National Laboratory for High Energy Physics
Oho-machi, Tsukuba-gun, Ibaraki-ken, 305 JAPAN
Introduction
TRISTAN is an electron-positron-proton colliding
beam facility of KEK (National Laboratory for High
Energy Physics) [I]. The construction starts this
spring and it takes five years before the commissioning
of the electron-positron collider of 30 x 30 GeV.
Then we will add a superconducting proton ring with a
maximum energy of 300 GeV for electron-proton or
positron-proton colliding. Its completion is expected
to be in late 80s.
The electron-positron collider consists of three
major parts: a 2.5 GeV linear accelerator, a 6 GeV
accumulator ring and a 30 GeV main ring. In this
report we present the design of the computer control
system of the accumulator and the main ring of TRISTAN.
Basic Design Concepts
1. Distributed Control TRISTAN is a very complicated
and large machine. It consists of a number of subaccelerators
and storage rings, which cover a large
area. Large number of equipments are distributed
around it. Complex operations are necessary in order
to fully utilize its ability. These complexity and
size of TRISTAN make it reasonable for us to adopt a
scheme of the distributed computer control system. We can point out the following advantages of the
distributed control system:
(1) The system is flexible, since we can modify
the subsystems without affecting the rest of the
system.
(2) Failures in one or more subsystems do not cause
the entire system down.
(3) cost of cabling is reduced due to short
distances between computers and controlled
devices.
(4) Number of transactions between subsystems are
reduced, since preprocessed data are transmitted.
2. Languages A complex accelerator such as TRISTAN
necessitates great software efforts, since elaborate
operation of it and thorough investigation of its
nature are possible only by means of good control
programs. Therefore, it is advisable that the device
designers and the operators, who know the algorithm
better than the other people, write control programs.
There lies a barrier in front of us, namely 'software
barrier' [2]. The difficulties in making application
programs are the main cause of preventing people from
being familiar with computers. In the field of the
distributed accelerator control this circumstances are
more severe, since there are many problems inherent to
the distribution of intelligences. For example, we
must take into account the synchronization of two or
more processes that run on different computers. This
is a difficult task, if we write programs using a
compiler language such as FORTRAN.
We have decided to adopt NODAL interpreter [3],
which was devised and has been successfully used in
CERN SPS, to overcome the above mentioned 'software
barrier'. The reasons for adopting NODAL are: I
(1) We can easily debug programs written in NODAL,
since it is an interpreter.
(2) NODAL has the clear idea as to incorporate device
handlers as data modules.
(3) NODAL has powerful string handling capabilities,
which are very useful when we write programs
that handle man-machine interfaces.
(4) In NODAL the lines are divided into groups;
separate groups of a program can be executed as
subprograms. This enables us to write
structured and readable programs.
(5) NODAL is a multi-computer language; it allows a
program to be expressed as a number of separate
tasks which can be executed on different
computers. This greatly simplifies programs
which run on a multi-computer environment.
(6) NODAL has powerful real-time facilities.
3. Interface Standard We adopt serial CAMAC as the
interface standard for the control of TRISTAN. The
reasons are:
(1) It is the only commercially available data way
which connects devices to computers over long
distances with a sufficient transmission speed.
(2) Bypa= and loop colapse functions of it are
useful for the system maintenance.
(3) The test of prototype devices is possible before
the commissioning of the control system, since
CAMAC can be connected to a computer that is
different from those used in the TRISTAN control.
Network
There are various techniques to link minicomputers
to form a local network. For example, in
SPS they use message handling computers based on the
store-and-forward technique [2]. Though it is a very
simple method, it has a drawback that the transmission
speed is relatively low. Moreover, if the number of
computers connected to the network is increased, we
must use more than two message handling computers.
This increases the complexity of the system and makes
the response slow.
We adopt the method which has become popular in
the industrial control such as large steel
manufacturing plants. All nodes are connected with
optical-fiber cables to form a ring network. The
transmission speed on the optical-fiber cables is 10
Mbps. When a node wants to transmit a message to
another node, it senses whether it can transmit the
message or not. If the line is free, it transmits the
message to the destination node. If the line is
occupied by other nodes, it must wait until1 the line
becomes free. The function of message switching is
0018-9499/81/o6oa-23S9$~.7501981 IEEE 2359
© 1981 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material
for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers
or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.
distributed among the- receiving nodes; each node
recognizes and accepts the message addressed to it.
There is no central node for switching. The
transmission schema is ,therefore, that of an N-to-N.
This method ensures the very effective transmission.
III our case overall transmission capacity is estimated
to be approximately 300 kwords/sec; this speed is
approximately 10 times faster than that of SPS.
Another advantage of this method is that addition of a
new node is easy and simple. We need only to cut the
cable and insert a new node. This is very convenient
for us, since the number of the nodes will be
increased as the development of the accelerator.
The network for the TRISTAN control system is
illustrated in Fig. Twenty-five similar 16-bit minicomputers
are linked to two loops. Four local
computers for the accumulator ring hardwares are
connected to the small loop, whereas fourteen
computers for main ring hardwares are connected to the
large one. Seven central computers, which are located
in the control center, are common to the both loops.
There exists only one control center for TRISTAN; the
facilities of central computers are commonly used for
the accumulator and the main ring controls.
Software System
We adopt the ideas of SPS concerning the software
system. These are:
1. Use of an interpreter language NODAL as a
control language.
2. Use of a distributed data base.
3. Use of data modules for device handlers.
4. Assignment of linkmen aa interfaces between
the control group and hardware groups.
We intend to speed up NODAL interpreter by the
following methods: (1) We make such NODAL interpreter
that is similar to NODILER in SPS, that is to say, an
incremental compiler [4]. (2) We use the
microprogramming technique to fix frequently used
portions of NODAL into firmwares.
Data modules are closely related to the hardwares
driven by them. Therefore, it is convenient to
develop data modules at the place where the hardwares
exist. To this end, every local computer is equipped
with a disk and a CRT terminal and can be used aa a
stand-alone program developer. Using it, we can make
and test data modules by examining the action of the
hardwares.
We develop NODAL compiler as a tool for making
data modules. It must generate efficient codes to
minimize the memory size and execution time, since they
are important factors for data modules. By the use of
a set of interpreter and compiler of the same language
specification, the efficiency of making data modules
will be greatly increased. We use the interpreter
first to utilize the easiness of debugging programs.
After debugging, we compile the programs to get object
codes.
Process Interfaces
We can distinguish three levels in our system.
The first level is the mini-computers and the links
between them. In our system, main computing powers
are concentrated in the mini-computers. The main
transactions among mini-computers are exchanges of
NODAL source codes. The average size of the messages iS estimated to be around ZOO-300 bytes. The second
level consists of CAMAC serial highways and CAMAC
2360
modules. From each local computer a CAMAC serial
highway is extended over the devices. Bit serial
CAMAC highway is adopted to minimize the cable cost and
to simplify the connections. The transmission speed
of it is 2 Mbps. The intelligence in the second level
is CAMAC auxiliary crate controllers. The main
purpose of them is to buffer data and adjust the
difference of speed between computers and devices.
The third one is the device level. If a closed
feedback loop is necessary, it must be incorporated in
the device. Transactions between mini-computers and
devices are, therefore, simple commands and data.
To summarize, the first level communication can be
regarded as the mean of communication between
intelligences, whereas the nature of the second level
point
is the
is
extension
essential
of computer
for the choice
input-output
of serial
buses.
CAMAC
This as
the process interface standard for TRISTAN. For
example, there is criticism that CAMAC is not a
suitable interface standard, because it is not devised
to be a mean to communicate between intelligent
devices. However, recent progress of micro-processors
and LSIs enbles us to make intelligent devices.
Control algorithms are fixed in the actual devices in
the form of codes in PROMS; complex intelligences in
CAMAC crates are not necessary. The nature of CAMAC
as an extension of the computer bus is adequate for
this kind of applications.
Control Center ~-
TRISTAN should be controlled and monitored from a
single control center. The functions necessary to
the control center are divided into four subfunctions
which are managed by functionally specified minicomputers.
They are:
Operators Console Function,
Alarm and Logging Function,
Program Library and Data Base Function, and
Program Development Function.
Man-machine interfaces are very essential to the
accelerator controls, since they are the means that
facilitate the communication between operators and the
accelerator. The basic philosophy of our man-machine
interfaces is to have general purpose consoles. The
improvements of the functions is achieved by increasing
the number of the general purpose consoles and refining
the control softwares. If we allow the use of a
special console, the motivation of developing softwares
to improve the controllability will be suppressed. At
the beginning the special purpose console may be
convenient, because it is made up for the special use.
When we want the more elaborate control of the
accelerator, this speciality limits the expansibility
of the control program.
Program library and data base are managed by the
file computer. It is similar to other computers
except that it is equipped with large volume disks.
All application programs, data modules are filed in the
disks. At the start up of the system, the necessary
files are sent from the file computer to the others and
are stored on local disks; it helps to reduce the
number of transactions on the network. Then data
modules are loaded from the local disk into the memory.
This reduces the overhead necessary for data movement
from the disk to the memory.
There are a few cases which cannot be manipulated
by the mini-computers on the network. Examples of
them are execution of the large size programs for
accelerator simulation and management of large data
base of the accelerator history. For this purpose we
use a back-end medium scale computer which is connected
to the network through the program development
computer. Also the central computers of our
laboratory, Hitachi M200Hs, are linked to the network
by means of a 48 kbps synchronous communication line.
From any computer in the network we can communicate
with them. Any terminal of the mini-computer can be
regarded as a TSS terminal of the MZOOHs.
Summary
We summarize here the outline of our system.
(1) Distributed computer control scheme is adopted.
(2) Twenty-five l&bit mini-computers are connected
by 10 Mbps optical fiber cables to form an
N-to-N ring network.
(3) NODAL interpreter and compiler are used
throughout the program development of the system.
(4) CAMAC serial highways are used as the means of
communication between the mini-computers and the
devices.
OPERATOR CONSOLE
(5) TRISTAN system is controlled from a single
control center.
Acknowledgements
We wish to thank Professors T.Nishikawa, S.Ozaki
and H.Takahashi for their valuable discussions.
References
1. T.Kamei and Y.Kimura in this proceedings.
2. M.C.Crowley-Milling CERN 75-20 and CERN 78-09.
3. M.C.Crowley-Milling and G.C.Shering CERN 78-07.
4. J.Altaber and P.D.Van de Stok
CERN SPS/ACC/PVdS/Rep. 79-4.
MEDIUM @,-*I c1
COMPUTE
1 5,y-j 11, iT (24, fi$
CONSOLE
lfACc~h4u~AToR RING NETWORK (loMbpi 1 MAIN RING NETWORK
FREQUENCY TRANSPORT
TRISTAN SYSTEM
2361
