Inter-process communication (ipc) has to be fast and e ective, otherwise programmers will not use remote procedure calls (RPC), multithreading and multitasking adequately. Thus ipc performance is vital for modern operating systems, especially µ-kernel based ones. Surprisingly, most µ-kernels exhibit poor ipc performance, typically requiring 100 µs for a short message transfer on a modern processor, running with 50 MHz clock rate. In contrast, we achieve 5 µs; a twentyfold improvement. This paper describes the methods and principles used, starting from the architectural design and going down to the coding level. There is no single trick to obtaining this high performance; rather, a synergetic approach in design and implementation on all levels is needed. The methods and their synergy are illustrated by applying them to a concrete example, the L3-kernel (an industrial-quality operating system in daily use at several hundred sites). The main ideas are to guide the complete kernel design by the ipc requirements, and to make heavy use of the concept of virtual address space inside the-kernel itself. As the L3 experiment shows, significant performance gains are possible: compared with Mach, they range from a factor of 22 (8-byte messages) to 3 (4-Kbyte messages). Although hardware specific details in uence both the design and implementation, these techniques are applicable to the whole class of conventional general

Eumel and its advanced successor L3 are operating systems built by GMD which have been used, for 13 years and 4 years respectively, as production systems in business and education. More than 2000 Eumel systems and 500 L3 systems have been shipped since 1979 and 1988. Both systems rely heavily on the paradigm of persistence (including fault-surviving persistence). Both data and processes, in principle all objects are persistent, files are implemented by means of persistent objects (not vice versa) etc. In addition to the principles and mechanisms of Eumel /L3, general and specific experiences are described: these relate to the design, implementation and maintenance of the systems over the last 13 years. For general purpose timesharing systems the idea is powerful and elegant, it can be efficiently implemented, but making a system really usable is hard work.