Page 2 of 3   « previous page   next page »

Evaluation of Differential QoS Support in Emerging Web Services Management Frameworks
So far there are no established standards for formally specifying QoS assurances and classes of service in machine-readable ways. However, various XML-based languages have been proposed such as Web Service Level Agreement (WSLA) and Web Services Offerings Language (WSOL). Such languages help define contracts to specify agreed-upon, nonfunctional characteristics of Web services (in the form of Service Level Objectives - SLOs) as well as a model for measuring, evaluating, and managing the compliance with these characteristics.

The WSLA framework consists of a flexible and extensible XML language for the specification of custom-made SLAs as well as a management infrastructure that comprises several SLA enforcement and monitoring services both for dynamic allocation of resources and for compliance checking.

WSOL enables the specification, monitoring, and management of classes of service for Web services, instead of custom-made SLAs. WSOL is the only framework that has an explicit notion of class of service. A class of service refers to a formal representation of a discrete variation of QoS guarantees provided by a Web service. WSOL also provides mechanisms for the specification of static and dynamic relationships between different classes of service. The latter are used in the algorithms and protocols to dynamically modify the QoS levels offered by a service. The corresponding Web Service Offering Infrastructure (WSOI) enables monitoring of Web services described in WSOL and implements the dynamic adaptation algorithms and protocols.

In terms of supporting negotiation of service level, WSOL only allows basic negotiation before switching to a different service level at runtime. The "autoManipulation" attribute determines whether the provider is allowed to perform switching to a different service offering without asking the consumer for explicit confirmation. If the value of this attribute is "True" the provider does not need to ask for consumer confirmation before switching. However, the provider has to inform the consumer after the switching. On the other hand, WSLA has more powerful negotiation capabilities that are further explored in the WS-Agreement specification.

For admission control, WSLA uses advanced control mechanisms that take into account the current load and available resources, whereas, WSOL uses simple access rights to specify conditions under which a consumer of a service offering has the right to invoke a particular operation. For example, an access right can limit the time of day when an operation can be invoked. It can also limit the number and/or frequency of invocations.

Inferences
WSLA was developed as part of IBM's autonomic computing initiative, and seems to be the most complete differential QoS framework. WSLA is extensible and allows logical expressions such as trade-offs between QoS parameters to be expressed; it also allows failure behavior to be specified. It defines a framework for providing differentiated levels of service through automated management of resources according to the negotiated SLAs. Some extended sets of WSLA abilities have been already integrated in IBM WebSphere Extended Deployment (XD). The WSLA project spawned into a number of efforts like WS-Agreement standardization in the Global Grid Forum. However WSLA is relatively more complex and introduces high negotiation and management overhead.

WSOL on the other hand is a simple and lightweight framework used to describe and manage multiple classes of service for a Web service, but it does not address how to provide the offered service levels.

See Table 1

The proposed approaches for expressing differential QoS assurances are still immature in terms of software implementation and experimental evaluation. Also they are silent when it comes to QoS provision, except for WSLA. Furthermore, the absence of a common QoS management standard might impede interoperability. The differential QoS space remains fragmented and a unifying framework is highly needed to tie up all of these approaches into a coherent framework. For example the distinctive features of WSOL could be added to WSAL and the two can be integrated with WS-Policy, with the latter serving a container for assertions.

The requirement of automated enforcement of machine-readable policies that control the variation of differentiated services has not been addressed in any of proposed solutions. However, IBM's Policy Manager for Autonomic Computing (PMAC) can be seen as a step towards this.

The biggest challenge facing the provision of differentiated services is the automated mapping of service offerings or SLA requirements to resources requirements across abstraction layers in order to determine the detailed resource configuration necessary to meet the promised QoS assurances. This requires accurate service performance modeling, capacity management, and timely information on available resources. The second challenge is the integration of differential QoS mechanisms that are available at different levels along the message path: network, application servers, and SOAP layer. These challenges are further complicated by the virtualization of services and the composition of Web services into added-value aggregate services. Grid and autonomic computing initiatives are the space to watch for potential answers to these challenges.

Page 2 of 3   « previous page   next page »