While it has long been suggested to use parallelism to boost the performances of telecommunication switching equipment, existing solutions do not easily scale up to very high throughput unless dedicated hardware is used. In this paper, we i n vestigate how oo-the-shelf distributed computers could still be used along with relevant s o f t ware architectures to build truly scalable telecommunication switching equipment, taking as an example the implementation of an SMDS (Switched Multi-megabits Data Service) server. We rst analyse performance gures of an experimental SMDS server, and deduce the hardware and software architecture that is needed to make it scal-able. We then describe a prototype implementation, and discuss performance results. We conclude on the interest and perspectives of this kind of scalable architecture.