A new linear program for QoS-aware web service composition based on complex workflow

In this article, we propose a new model based on mixed linear programming to determine a composite web service (structured by a workflow) minimizing a QoS measure while satisfying some QoS constraints. The proposed mixed linear program is solved using a standard solver (CPLEX). Our experiments show that big-size instances can be exactly solved. To the best of our knowledge, it is the first time that a linear program with a polynomial number of variables and constraints is proposed for QoS-driven web service composition based on complex workflow with interlaced patterns. I. WEB SERVICE COMPOSITION Web services (WS) are the most famous implementation of service oriented architectures (SOA) [1] allowing the construction and the sharing of independant and autonomous softwares [2]. WS composition consists in combining several existing WS into a composite one, which becomes a valueadded process [3]. The WS composition problem aims at identifying a set of existing WS such that the composition of those WS can satisfy the user’s functional and non-functional requirements [4]. To differentiate several WS having the same functionality, QoS criteria can be used to select the ”best” WS satisfying the users’ requirements. The following sub-sections describe our context: the workflow structure of the composite WS and the QoS criteria. Then, some related works are studied. Section II presents and justifies our new mixed linear programming model. Section III presents some very promising experimental results that show the relevance of our model. Finally, Section IV concludes. A. Workflow A workflow describes how to combine the functionalities of different WS in order to satisfy the user. In a workflow, an activity represents a set of WS sharing a same functionality, and a pattern represents temporal dependency between different activities. In this article, we consider three patterns: sequence, parallel (AND) and exclusive choice (XOR). Figure 1 represents these workflow patterns, based on YAWL model [5], where Ai is an activity. The sequential pattern, see Figure 1.(a), indicates that A1 must be executed before A2. The XOR and AND patterns start with a split and finish with a join. In AND pattern, see Figure 1.(b), all activities A1, . . . , Ak have to be executed in parallel. For XOR pattern, see Figure 1.(c), only one activity among A1 to Ak has to be executed. In the following, we consider general complex workflows in which these patterns can be recursively concatened and (a) Sequential pattern (b) AND pattern (c) XOR pattern XOR-join AND-join XOR-split AND-split ... ... A1