A Pattern Language For Improving the Capacity of Layered Client/Server Systems with Multi-Threaded Servers

The paper describes a set of patterns that extend the pattern language proposed in [Meszaros96] for improving the capacity of reactive systems. The intent of these patterns is to identify some specific causes that limit the efficiency of a distributed layered client-server system with multi-threaded servers, and to find appropriate corrective measures. The type of systems considered here is a subclass of the larger category of reactive systems, and the new patterns are dealing with their specific performance characteristics. The effects of the patterns are illustrated with performance measurements conducted on a layered client-server system. INTRODUCTION Problem Domain Many distributed applications are based on the client-sever paradigm and use various kinds of software servers (as for example, name servers, databases, network file servers, web servers, etc.) The performance of such systems depends strongly not only on the contention and queueing delays for hardware devices (such as ...