Business transactions in web service environments run with relaxed isolation and atomicity property. In such environments, transactions can commit and roll back independently on each other. Transaction management has to reflect this issue and address the problems which result for example from concurrent access to web service resources and data. In this paper we propose an extension to the WS-Transaction Protocol which ensures the consistency of the data when independent business transactions access the data concurrently under the relaxed transaction properties. Our extension is based on transaction dependency graphs maintained at the service provider side. We have implemented such a protocol on top of WS-Transaction. The extension on the web service provider side is simple to achieve as it can be an integral part of the service invocation mechanism. It has also an advantage from an engineering point of view as it does not change the way consumers or clients of web services have to be programmed. Furthermore, it avoids direct communication between transaction coordinators which preserves security by keeping the information about business transactions restricted to the coordinators which are responsible for them. 1.

For improved flexibility and concurrent usage existing transaction management models for Web services relax the isolation property of Web service-based transactions. Correctness of the concurrent execution then has to be ensured by commit order-preserving transaction schedulers. However, local schedulers of service providers typically do take into account neither time constraints for committing the whole transaction, nor the individual services ' constraints when scheduling decisions are made. This often leads to an unnecessary blocking of transactions by (possibly long-running) others. In this paper, we propose a novel nonblocking scheduling mechanism that is used prior to the actual service invocations. Its aim is to reach an agreement between the client and all participating providers on what transaction processing times have to be expected, accepted, and guaranteed. This enables service consumers to find a set of best suited providers fitting their deadlines. Service providers on the other hand can benefit from the proposed mechanism due to the now possible intelligent scheduling of service invocations for best throughput. In fact, our experiments show a significant improvement in terms of overall throughput, service chain completions and resources ' utilization. 1.

apport de recherche

apport de recherche