Microsoft word - houston.doc

AMS : An Integrated Simulator for Open Systems
Atika COHEN and Radouane MRABET
Bd du Triomphe, CP 230 Brussels, Belgium cohen@helios.iihe.rtt.be and mrabet@helios.iihe.rtt.be C=be; ADMD=rtt; PRMD=iihe; O=helios; S=cohen or S=mrabet
ABSTRACT The main objective of the OSISIM
Network Analysis Package 2) and GSS4 [8] project (Open System Integrated Simulator) is to set up (Graphical Support System 4)1 are the main an atelier for the modelization of communication software packages that will be used to operate the networks and the analysis of their performances. The atelier gives the end-user powerful tools to edit a communication system in a graphical environment. The One component of this atelier is a library of basic representation of a system to be simulated is based on models including most of the standard networks models of several standard networks available in the such as LANs (Ethernet, Token Ring, FDDI, etc), library which is the kernel of the atelier. Each model implements uniform and well-defined sets networks. It is obvious that the usefulness of the of functions, while having clearly specified interfaces. This article describes the architecture of the atelier, atelier heavily depends on the number of and focuses on the internal structure of basic models. Introduction
The end-user of the AMS is not expected to be a The main objective of the OSISIM project is to specialist in modeling or performance analysis; set up an atelier for the modelization of however, he or she should be a communication communication networks and the analysis of their system designer. He or she will use the AMS to performances. In the literature, we find the build and validate an architectural choice, or to description of different toolkits dedicated to this compare several possible ways of solving a field, as TOPNET [1], which is based on PROT net, a class of Petri nets; NETMOD [2], which is Models have to be constructed in a very modular based on simple analytical models; BONeS [3], fashion. That is why we have to build basic which is based on block-oriented modeling models, which will make up other models of more paradigm. On the other hand, our approach is complex systems. Each basic model implements uniform and well-defined sets of functions, while Modelization and Simulation), has to integrate the This paper describes, in the first section, the facilities and the tools in such a way as to be easy architecture of the AMS [4]. The second section to use by the end-user. It gives the end-user some focuses on the internal structure of a basic model [5], and the ways to reduce the code related to its • basic library that includes the models of several standard networks studied separately; I. Description of AMS
• tools for editing communication systems and I.1. AMS architecture
• tools for visualizing simulation results; Figure 1 shows the architecture of AMS. The components of this architecture are arranged in The AMS will be based on the latest techniques three principal groups : the objects that can be in software engineering such as : graphics, windowing, pop-up menus and the object-oriented programming paradigm. QNAP2 [6,7] (Queueing These packages are trademarks of SIMULOG, a French company specialized in the field of simulation. manipulated by end-users (i.e. basic models) in b) Scenario_Def : This process helps the end-user
order to describe the communication system to be simulated, the processes for editing and c) Result_View : This process allows the end-
generating code, and the set of files containing the user to choose the types of simulation results he internal representation of the system under study. Scenario_Element
d) Code_Generator : This process generates the
Draw_Element
code that describes the modelization of the global system by means of the elements edited by the Arch_Draw
Scenario_Def
end-user. This code is mainly written in the Scenario
QNAP2 language. The efficiency of this code Result_Element
depends upon the internal structure of the basic models. Before the code is generated, the validity Result_View
Basic Models
e) Processor : This process compiles and
executes the generated code. It requires the description of the simulation experiment which includes a number of data for simulation control Code_Generator
I.1.3 Files containing the description of the system Data + Simulation
Processor
a) Arch_Descrip : A file which contains the
description of the system edited. This file is b) Scenario_Descrip : This file contains the
figure 1 : The architecture of the AMS functional description of the simulation steps. It is Hereafter follows a description of these c) Result_Descrip : This file contains the
description of the results. It is generated by the I.1.1 Objects manipulated by the user a) Basic Models : The core of the AMS is the
d) Simul_Descrip : This file contains code that
library of basic models. A basic model is mainly a describes the modelization of the whole system. It communication entity such as a standard network, is generated by the Code_Generator process. protocol, gateway, etc. Each basic model is e) Results : This file corresponds to simulation
characterized by a definite number of parameters results The results are presented under the form and an exhaustive list of measurements; in fact, In order to facilitate the use of these models, their I.2. Functional description of AMS
interconnection and their maintenance, a unified The AMS is composed of four processes which internal architecture for basic models is defined are Arch_Draw, Scenario_Def, Result_View and (see the following section) and used to structure Code_Generator. These processes, except the last one, execute their codes in parallel. They b) Draw_Elements : Design objects for basic
c) Scenario_Elements : These objects define the
scenario which will be followed during the Let us describe how an end-user works with the AMS. First of all, he executes the Arch_Draw d) Result_Elements : These objects define the
type and the form of results the end-user wants to Draw_Element, he prepares the system to be modeled. For each basic model, he can modify the predefined values of parameters while taking into account the variation limits of the parameters. a) Arch_Draw : It is the process that allows the
With regard to measurements, the end-user can end-user to edit his system graphically. It generates the description of the edited system. measurements, or, for specific measurements he The validity of the system is checked during the edition such as the connectivity between basic When the description of the system is complete, he executes the Scenario_Def process, and uses Scenario_Element to elaborate different scenarios based on the parameters of the basic models. He also executes the Result_View process in order to determine the results he wants in accordance with The end-user can call these processes in any order. The process Code_Generator can be called only if all the elements required to generate the figure 2 : Internal structure of a DBM code A DBM can have several interfaces, N in figure 2. Calling the Code_Generator process blocks up the That number N can either be a fixed value known three other processes. It is going to validate the during the modelization phase (for instance, an different editions done by the end-user. If the optical fiber cable can connect two workstations process detects some contradiction or omission, it and therefore N=2); or N can vary so that the signals to the end-user what the problem is, stops modeler can specify only the minimal and the its execution, and releases the three other maximal values (for instance, an Ethernet cable processes, in order to allow the end-user to The validity of the interconnection between To launch the simulation the end-user has to themselves but with tools belonging to the AMS. II. Basic Models
Figure 3 shows how an interface, using messages, reacts with its BE and with the outside world (i.e. another interface belonging to another DBM). II.1 Internal structure of basic models
Each basic model is to be detailed so as to reflect
its exact behavior. Hence, we have to specify the functions performed by the basic model as exactly as possible. Hereafter, basic models will be called Detailed Basic Models (DBMs). The primary advantage of this approach is to have accurate measurements and to highlight the largest possible number of parameters to characterize the As a rule, a DBM can't be used alone, it must be figure 3 : A DBM with its BE and one Interface connected with one or several other DBMs, in Figure 4 represents a standard interface with its order to make up a complex system. That's why two internal queues. Qio receives messages from each DBM needs one or more interfaces so as to the outside, to be sent later to the BE. Qii receives messages from the BE to be sent later to another Figure 2 shows the structure of a DBM code. There are three blocks : Behavior Engine (BE), Interfaces (Int) and Measurement Block (MB). In the BE block, we find a modelization of the behavior of the DBM. The behavior is controlled by a set of parameters so that the end-user can choose the values he wants to allocate to each Seen from the outside, all the interfaces are similar but they differ in the way they interact with their BEs. An interface doesn't play any active role, its main function is to convey messages from a BE to the outside and vice versa. Each DBM has an identifier on which we find the exhaustive list of parameters and their default values, the measurement list, and a few text lines, summarizing the main functions of the DBM. • Unless deleting a part of the behavior entails deleting the measurements associated with that II. 2 Reduced Basic Models (RBM)
part, the end-user is not aware of the existence Each basic model is to be detailed so as to reflect of RBMs and DBMs, all that he knows, is that its exact behavior. Namely, the functions performed by the basic model are to be specified • The same DBM code can be broken down into as exactly as possible. The reasons why we have several RBMs, depending on which part is to be deleted. That is why the DBM code has to • to have a model behavior close to a real • to derive accurate measurements from the For the deletion to be efficient, the DBM code must be written in such a way as to facilitate this However, this approach has some drawbacks task. In other words, the different parts of the because of the large size of the code resulting • the compilation is very time consuming; II.2.2 Logical y reducing the size of the code • the simulation is also very time consuming; With this method, no part of the DBM's code is • the code describing a system is, of course, very deleted physically but some parts of it will not be executed when the simulation is running. several DBMs, each of them having a long The model builder (modeler) adds a set of • during the simulation run, a large-sized predicates in the DBM's code. Each predicate governs a part of the code and when this predicate • the end-user will be submerged with details, so is true, the code associated with it will be that his work for editing will be difficult. executed, whereas if this predicate is false the In order to soften the effects of these drawbacks, Each predicate is a combination of elementary eliminating, for example less significant details. conditions : let C=(c1, c2,.,cn) be the elementary The code so obtained will be called RBM for conditions vector, with ci equal to 1 or 0, and Reduced Basic Model. To do this, we are faced P=(p1, p2,., pm) the predicate vector with pi=f(C), i=1, 2,., m, fi is a logical function using • how can less significant details be chosen ? operators such as AND, OR, NOT, etc. We can • during the construction of the system code, the have 2n different vectors C and k vectors P with question that arises is : for which component is k≤ 2min(n,m). We can have k different versions of the DBM required ? and for which one is the RBM sufficient ?; the choice may be made II.2.3. Using assumptions to simplify the DBM even more difficult, because each DBM can The third method consists of simplifying the The primary drawback of reducing DBMs is that DBM's code by making assumptions aimed at the simulation measurements are less accurate, reducing the complexity of some algorithms because the RBMs do not reflect the whole which describe the behavior. This can be done by behavior of the real system. An RBM has the physically replacing the complex algorithms with There are several methods to reduce a DBM code, II.2.4. The problems which arise when we want to II.2.1. Physical y reducing the size of the code The following choices have to be made by a competent modeler who has to decide which part In order to reduce the size of a code, each DBM of code is to be deleted, which algorithm is to be will have several parts of its code deleted replaced, etc. Besides, he has to determine which physically. These parts may represent more or parts of the DBM's code significantly influence less significant parts of the whole behavior, yet, deleting them does not significantly change the behavior, in so far as the remaining code is still To achieve this, the modeler can be helped in three ways : either by experts in the field of network communications; or by researchers' Conclusion
theoretical and experimental studies; or he can The paper describes briefly the atelier AMS, make a simulation, called local simulation, for a designed to evaluate the performance of open specific DBM, possibly interconnecting it to the systems. It comprises a library of basic models. minimum number of DBMs required to have a An internal structure is proposed for these models. The description of a basic model have to The modeler will often use the hybrid method be reduced, in order to diminish the impact of (when several methods are used to reduced one DBM) because it gives him much flexibility, but memory size. These drawbacks appear when the this flexibility involves building several RBMs system under study is more or less complex. from the same DBM. This proliferation of RBMs entails several problems which are fairly hard to References
resolve. The first problem can be phrased as [1] M. A. Marsan, G. Balbo, G. Bruno, F. Neri, follows : "Given a DBM to be reduced, which "TOPNET : A Tool for the Visual Simulation of Communication Networks". IEEE Journal on Selected Areas in Comm., Vol. 8, no 9, 1735-1747, Dec. 90. II.3. Generic Models (GeMs)
[2] D. W. Bachmann, M. E. Segal, M. M. Srinivasan, and T. J. Teorey, "NetMod: A Design Tool for Large- In order to classify basic models, we define what Scale Heterogeneous Campus Networks", IEEE Journal we call the Generic Models (GeMs), which are on Selected Areas in Comm., Vol. 9, no 1, 1735-1747, model classes. Each Basic model belongs to one or several GeMs according to the functionalities [3] K. S. Shanmugan, V. S. Frost, W. LaRue, "A block- which it handles. This classification will allow us Oriented Network Simulator", Simulation, 83-94, to use a tool to verify the validity of a system [4] A. Cohen and R. Mrabet, "AMS : Atelier for The classification of the basic models will be done on the basis of several criteria, for instance, [5] A. Cohen and R. Mrabet, "AMS : Internal structure the largest possible number of links between one for Basic Models", Internal Report, IIHE/HELIOS-B- basic model and the other ones; the OSI stack [6] "QNAP2 User's Manual", Simulog S.A. 1992. layer a basic model belongs to; the fact that a [7] "QNAP2 Reference Manual", Simulog S.A. 1992. basic model is terminal or not, namely generating [8] "GSS4 User's Guide", Simulog S.A. 1992. Acknowledgments
II.4. Local and Global Measurements
We wish to express our gratitude to SAIT Electronics with whom we are colaborating on the OSISIM project. We also gratefully acknowledge measurements, called local measurements (LMs). A default list is also defined for each basic model. In general, the defaults lists are defined in the same manner for all basic models and contain typical measurements which are related to the information handled by basic models or to the use Although typical measurements are defined in the same way, they are processed differently because basic models behave in different manners. The end-user is not only interested in LMs, but also in measurements related to the whole system which (s)he has edited. These measurements are called the global measurements (GMs). The AMS gives the end-user the means to define a GM in function of LMs, but the semantic of the GMs has to be defined by the end-user, except for some verifications which are done by the AMS in

Source: http://www.ensias.ma/ens/mrabet/houston93.pdf