Deriving protocol specifications from service specifications written as Predicate/Transition-nets

Abstract

We consider the derivation of a protocol specification from a service specification written in Predicate/Transition-nets (Pr/T-nets). The service specification describes the global behavior of a system and includes the allocation of the Pr/T-net places to N distributed sites. The paper presents a new algorithm for deriving a protocol specification that defines the behavior of N communicating entities that execute on the N sites and coordinate their actions in order to conform to the global behavior defined by the service specification. Our algorithm decomposes each transition of the service specification into a set of communicating Pr/T-subnets running on the N entities. Moreover, for efficiently controlling the conflict for shared resources, we present a timestamp-based contention control algorithm and incorporate it into the derivation algorithm. A tool has been developed that implements our algorithm and works together with other existing tools for the graphical representation of the service and derived protocol specifications. Two application examples are discussed.

Introduction

Synthesis methods have been used to derive the specification of a set of application components running in a distributed system of networked computers (hereafter called protocol specification) automatically from a given specification of services to be provided by the distributed application to its users (called service specification). The service specification is written in the form of a centralized model, and does not contain any message exchanges between different physical locations. However, the definition of the behavior of the application components, called protocol entities (PEs), includes the message exchanges between these entities. Protocol synthesis methods have been used to specify and derive such complex message exchanges automatically in order to reduce the design costs and errors that may occur when manual methods are used.

Many synthesis methods have been proposed in the literature. The methods use different computational models as service definition languages. For example, the methods presented in [1], [2], [3] use CCS/LOTOS models, the methods in [4], [5], [6], [7], [8] use FSM/EFSM models and the methods in [9], [10], [11], [12], [13], [14], [15], [16] use Petri net models. Similar methods may also be used for deriving distributed testers for distributed applications [17] and for deriving specifications of real-time systems [18], [19], [20]. In this paper we consider service and protocol specifications written in high-level Petri nets. These are extended Petri nets where tokens have values and the firability of transitions may depend on those values. Popular versions of high-level Petri nets are predicate/transition nets (Pr/T-nets) [21], [22] and coloured Petri nets (CPN) [23]. These models have enough modeling power, analytical power and tool support (such as CPN Tools [24]) to specify, verify and analyze large and practical software systems [25], communication protocols [26], [27], control systems and so on [23], [28].

In this paper, we propose a new algorithm for the derivation of a protocol specification in Pr/T-nets, which is the specification of N communicating entities (N is given), from a given service specification in Pr/T-nets and an allocation of the places of the service specification to the N entities. Our algorithm decomposes each transition of the service specification into a set of communicating Pr/T-subnets running on the N entities. Moreover, in order to improve the efficiency of controlling the conflict between different transitions over shared resources, we present a timestamp-based contention control algorithm and incorporate it into the derivation algorithm. A tool has been developed that includes our derivation algorithm and works together with other existing tools for the representation of the service and the derived protocol specifications. As application examples we discuss the application of our synthesis method to a distributed media transcoding service on overlay networks and to a distributed software development process [29].

Our approach is very powerful in the sense that general Pr/T-nets are allowed to be used for specifying services. Such Pr/T-nets may include complex conflict structures between transitions that require read and/or write access to shared resources in the form of tokens with values stored at shared places. Since these resources may reside on different sites and the transitions should be initiated when all required resources are available, we have to deal with this complex problem of distributed synchronization for the different transitions involving the protocol entities on the different sites. We first present a basic transition execution protocol where a transition is initiated by its “primary site” without having full knowledge about the available resources; the transition is then canceled whenever there appears to be some conflict or deadlock possibility.

Some existing synthesis methods also allow to treat variables (parameters) in their modeling languages as for instance a CCS-based model with I/O parameters [1] and Petri nets with external variables [9], [15]. However, since these existing methods mainly focus on value exchanges between entities, only simple control flows are allowed; the combination of choices and synchronization involving parameters, which often represents resource conflict, is not treated by those methods. Therefore, the class of acceptable service specifications has been considerably extended by the approach described in this paper. As far as we know, no previous paper has presented synthesis approaches for general Pr/T-nets.

We note that the basic idea of this paper was presented in [30]. Here we extend that work in several ways. First, we enhance the derivation algorithm by including a new derivation policy. Using a few additional messages, this policy prevents large size resources from being exchanged between entities. Second, we include a detailed timestamp-based contention control algorithm for efficiently controlling conflict for shared resources. Third, we developed a tool that includes our derivation algorithm and can interwork with other Petri net tools. Fourth, we provide arguments for the validity of our method and discuss the application of the method to two realistic examples.

This paper is organized as follows. Section 2 includes the definition of Pr/T-nets and provides examples of service and protocol specifications written in this notation. In Section 3 we present our derivation algorithm and in Section 4 we enhance this algorithm by incorporating a timestamp-based contention control algorithm. In Section 5 we describe two application examples and in Section 6 we conclude the paper.