1 WSPDS: Web Services Peer-to-peer Discovery Servicey Farnoush Banaei-Kashani, Ching-Chien Chen, and Cyrus Shahabi Computer Science Department, University of Southern California, Los Angeles, California 90089 [banaeika,chingchc,shahabi]@usc.edu Abstract| The Web Services infrastructure is a dis- One could anticipate popularity of this infrastruc- tributed computing environment for service-sharing. In this ture in advance, because it is an extension of the environment, resource discovery is required as a primitive functionality for users to be able to locate the services, the successful browser-based web-programming tech- shared resources. A discovery service with centralized ar- nology to a general distributed application devel- chitecture, such as UDDI, restricts the scalability of this opment environment. However, more importantly, environment as it grows to the scales comparable with the success of this infrastructure must be attributed to size of the web itself. In addition, current extensively used web service standards (e.g. UDDI, WSDL), do not support its fundamental features: discovery at a semantic level. In this paper, we introduce ² Loose coupling: services developed and deployed WSPDS (Web Services Peer-to-peer Discovery Service), a independently using heterogeneous platforms can fully decentralized and interoperable discovery service with semantic-level matching capability. We believe the peer-to- be integrated seamlessly to build distributed appli- peer architecture of the semantic-enabled WSPDS not only cations with new functionalities; hence, interoper- satis¯es the design requirements for e±cient and accurate ability. Loose coupling is mainly enabled by XML- discovery in distributed environments, but also is compati- based SOAP communication speci¯cation, which ble with the nature of the Web Services environment as a self-organized federations of peer service-providers without allows platform-independent information exchange any particular sponsor. between services. Keywords| Web Services discovery, Peer-to-peer ² Full decentralization: all communications of the discovery, Ontology, Semantic matching interacting entities are in a peer-to-peer fashion, without any central coordination; hence, scalabil- I. Introduction ity. The Web Services programming infrastructure ² Semantic level search: this feature allows web is the current generation of a succession of sys- service requesters to search for published web ser- tems proposed to develop distributed applications: vices not only based on keywords, but also based RPC, CORBA, DCOM, and now Web Services. A on ontological concepts. web service is a self-contained application module A. Discovery Service for Web Services with well-described functionality that can be in- voked across the web. The Web Services program- In general, in a distributed computing system a ming environment is a distributed computing envi- discovery service locates (or discovers) resources ronment in which participants share their services; dispersed across the system in response to re- hence, a service-sharing environment. Each partic- source discovery queries issued by the system enti- ipant can potentially act both as a service provider ties. With Web Services, resources are the services and as a client. As a service provider, the partic- shared on the web. To be speci¯c, a discovery ser- ipant builds and optionally shares its services for vice for Web Services is itself a web service that public use. As a client, on the other hand, the locates the service description document(s) of the participant can develop distributed applications by service(s) that hit a service query. A service de- discovery and seamless integration of the public ser- scription document (e.g., a WSDL ¯le) provides vices with its own private services. both abstract and concrete information required The Web Services infrastructure is adopted more for proper invocation of a service. A service query rapidly and widely as compared to its predecessors. characterizes a set of services with particular char- acteristics, such as name, abstract(or description), y This research has been funded in part by NSF grants interface model, etc., to be located. EEC-9529152 (IMSC ERC), IIS-0082826 (ITR), IIS-0238560 (CAREER), IIS-0324955 (ITR) and IIS-0307908, and unre- B. Design Issues and Approaches for Discovery stricted cash gifts from Okawa Foundation and Microsoft. Any opinions, ¯ndings, and conclusions or recommendations Service expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science To be compatible with the fundamental features Foundation. of the Web Services infrastructure (as discussed ion. Entities are peers in functionality and each entity is potentially both a server and a client of the peer-to-peer service; hence, sometimes entities with User with Communication are referred to as servents (i.e., server and client). The Web Services discovery service can be im- Communication User Query User t Communication with Neighbor n Engine Neighbor Query e v r Response to Neighbor Query e Response to User Query User to Response plemented as a peer-to-peer service, eliminating de- S S D P S W pendency on a distinct service provider. Each ser- A Local Query vent serves others by providing information about Engine its own web services in response to queries, and in turn, as a client it issues discovery queries to locate the web services that are not available locally. Ser- vents build a network in which each servent has a Local Inspection Documents few other servents as neighbors. When a servent re- Fig. 1. WSPDS Architecture ceives a request for a web service from the local user and cannot ¯nd the web service locally, as a client it originates a discovery query and propagates the re- quest into the network through its neighbors. Ser- above), a discovery service should support the fol- vents collaborate based on a distributed algorithm lowing requirements: to disseminate the query. During propagation of ² Interoperablity, to be integrable with other web the query, if a servent ¯nds the requested web ser- services, to support di®erent service description vice locally, it responds to the originator by pro- standards, and to be portable to di®erent plat- viding its location and description. forms; Additionally, in order to achieve e±cient query ² Scalability, to grow to the web scales without be- propagation in a peer-to-peer environment, the ing a performance bottleneck; linkage between servents should be built based on ² E±ciency, to support the dynamic environment the hosted data contents (e.g., web service descrip- of the Web Services with frequent changes/updates tions) of the servents. Finally, a more accurate of the location of the services and their description match will be accomplished by annotating both the documents; advertised web services and users' requests with ² Fault tolerance, to be resistant to unwanted globally shared concepts. breakdowns and malicious attacks. ² Semantic based discovery, to ¯nd a match based II. Peer-to-peer Discovery Service for on the common conceptual space of service re- Web Services questers and providers. The Web Services infrastructure is a self- We argue that as compared to a centralized ar- organized federation of service providers for service- chitecture (e.g., UDDI [1], the currently used stan- sharing. Thus, a peer-to-peer architecture is an dard for globally publishing and locating web ser- appropriate choice for the discovery service in this vices), a decentralized design for the Web Services environment. Considering the usual autonomous discovery service is more scalable (obviously), more behavior of the service providers, an unstructured fault tolerant (by eliminating the single point of peer-to-peer discovery service is preferred. Here, failure), and more e±cient (by reducing the over- we introduce WSPDS (Web Services Peer-to-peer head of centralized update of the discovery service). Discovery Service), a fully decentralized and in- Distributed directory services and peer-to-peer ser- teroperable discovery service with an unstructured vices are two alternative service models with decen- peer-to-peer architecture. tralized architecture. Distributed directory servers are usually dedicated facilities that are built and A. Architecture maintained under unique management to provide service to the clients of a distributed environment. WSPDS is a distributed discovery service im- However, the Web Services infrastructure is a self- plemented as a cooperative service. A network organized federation of peer entities without any of WSPDS servents collaborate to resolve discov- particular sponsor for the system. It is desirable ery queries raised by their peers. Figure 1 depicts that the federation lives, changes, and expands an unstructured peer-to-peer network of WSPDS independent of any distinct service facility with servents. Each servent is composed of two en- global authority. With peer-to-peer services, the gines, communication engine and local query en- role of distinct service providers is eliminated. Sys- gine, standing for the two roles that a servent plays: tem entities all cooperate to provide a service as a 1. Communication and Collaboration: the commu- result of group collaboration in a distributed fash- nication engine provides the interface to user and also represents the servent in the peer-to-peer net- erative discovery. These mechanisms are mostly work of servents. This engine is responsible for the compatible