WEB SERVICES MANAGEMENT NETWORK An Overlay Network for Federated Service Management Vijay Machiraju, Akhil Sahai, Aad van Moorsel Hewleff-Packard Laboratories IS01 Page Mill Road, Palo Alto, CA 94034 (vijaym, asahai, aad}@hpl.hp.com Abstract: We introduce the architecture, object model, components, and protocols of a management overlay for federated service management, called Web Services Management Network (WSMN). WSMN targets management of web services that interact across administrative domains, and therefore typically involves multiple stakeholders (examples are business-to-business, service provider interconnections, help desks). The architecture is based on (implicit) SLAs to formalize relations across domains. It relies on a network of communicating service intermediaries, each such intermediary being a proxy positioned between the service and the outside world. WSMN also exchanges control information to agree on what to monitor, where to monitor, and whom to provide visibility. Key words: management, service management, SLA, web services, web service management network 1. INTRODUCTION By packaging software applications as ‘services’ that are accessible over the Internet or intranet, enterprises achieve new and better means to utilize their own and each other’s applications. Services’ can be accessed through manual user activities (e.g., browser-based IT help desk), and increasingly through other services. In the latter case, conglomerations of interacting services emerge, which ’ We use the terms ‘service’ and ‘web service’ interchangeably, but prefer the term ‘service,’ since it stresses that we discuss the management of applications exposed as services, instead of the fact that we assume these services to communicate through web services technology (SOAP, XML, WSDL).352 Vijay Machiraju, Akhil Sahai, Aad van Moorsel access one another through more and more automated means. Examples can be found in business-to-business computing (web services, supply-chain processes, payment services), service providers (utility or grid computing) or in enterprise applications (payroll applications, remote IT services). In the emerging world of lntemet services, operational management becomes exceedingly important and challenging (and thus interesting). Services often directly impact the business process execution, and mishaps may directly be reflected in the bottom line of a business. This puts a premium on fault and performance management capabilities for services. Moreover, the services paradigm increases the complexity of run-time operations. Services communicate across monitoring domains [I], include different business partners, and are likely to rely on third parties to complete a service offering. As a consequence, service management has to deal with multi-party interactions, has to collect a large amount of data and synthesize it to understand the health of relationships between partners, and must resolve the limited end-to-end visibility and control one has over each other’s services. From the above we conclude that traditional application management must evolve into ‘true’ service management, and ultimately into service ‘relationship’ management. First, we must manage a service ‘as a whole,’ that is, as it is provided to a partner. Contrast this with application management, for which it is necessary to understand how the service is internally implemented through a set of objects and what exceptions each object generates. Instead, ‘true’ service management manages the interactions a service has with other senices or consumers, for which we need visibility at the service interface between an application and its users. Secondly, we must manage relations, not only through local management, but also through sharing data between partners as needed, and, more importantly, through exchanging signaling information about monitoring, se&ce levels and control actions. In other words, we need to be concemed with federated management [2]. In this paper, therefore, we propose Web Service Management Network, a management architecture for federated service management. Since we believe it is safe to assume Intemet services will he implemented using web service technology (SOAP, XML, WSDL), WSMN is based on such technology. The critical concept in WSMN is that of SLAs. If SLAs are explicitly defined we make them manageable, and if no SLAs are actually agreed upon between serviccs, WSMN manages towards SLAs imposed specifically for management purposes (‘implicit SLAs’). The SLA concept allows us to frame and solve many problems rather elegantly and effectively, as we discuss further in Section 2.1, and illustrate using a WSMN prototype implementation in Section 4. WSMN, then, is (I) a network of cooperating intermediaries, each such intermediary implemented as a proxy sitting between a service and the outside world, and (2) a set of protocols to manage service relationships expressed through SLAs. One can regard WSMN as a logical signaling network for management purposes, a concept well known from telecom (SS7 [3]). In that sense it is quite different from traditional management protocols such as those supporting SNMP and CIM based monitoring. It is closer to various overlays that are emerging throughout the various layers of the Intemet stack, to establish quality guarantees that the Intemet stackWeb Services Management Network 353 alone cannot create. Examples exist for instance for Internet telephony (SIP [4]) and streaming media content delivery [5]. Also of interest are the various emerging solutions to provide properties and features such as security, transactionality and change management to business-to-business web services. Examples are Flamenco Networks [6], Kenamea [7] and Talking Blocks [8]. All these companies use networks of collaborating intermediaries, often including a third-party play (repositories as well as services). However, none of these solutions addresses service management, as we do in this paper. Recently, Gartner surveyed and put in perspective these architectures, which they term web service networks [9]. Our term ‘WSM” is therefore extra appropriate, since our approach uses
View Full Document
Unlocking...