DEMOS/MP: The development of a distributed operating system

Abstract

The DEMOS/MP operating system has moved from a supercomputer with a simple addressing structure to a network of microcomputers. This transformation was done without significant changes to the semantics of the original DEMOS, i.e. existing DEMOS programs should run on DEMOS/MP. The changes to DEMOS were simplified by the structure of its primitive objects and the functions over those objects. The structure of DEMOS links and processes were the major contributors to the simplicity. The changes made to produce DEMOS/MP involved the internal structure of link, modification to parts of the kernel, and limited changes to the various system processes.

Process-originated migration in a heterogeneous environment

“…They include Eden, [ Needham [NEED82] reports work on migration in the cambridge Distributed System. Migration is also reported in DEMOS/MP, [MILL87] [POWE83] although that scheme leaves residual link pointers on the machine from where a program migrated. Locus and Eden [BLAC85] also provide migration.…”

Experience with Charlotte: simplicity and function in a distributed operating system

“…Second, the kernel at a receiving end sometimes needs to know the location of the sending end(s), to warn them, for example, that the receiving end has moved or been destroyed. In Arachne, Demos [9], and DemosIMP [27], all of which use unidirectional links, the information stored at the receiving end of a link is not enough to find the sending ends. Bidirectional links offer the opportunity to maintain information at both ends about the location of the other.…”

On the adequate support of communication interfaces in distributed systems

“…The communication model is supported at run time by a kernel which offers a suitable set of communication primitives implementing the model. These primitives are made accessible to users either by the provision of library interface packages as in Demos-MP [MIL87], Locus [WAL83], Amoeba [MUL84], and the V-kernel [CHE84] or by a distributed programming language as for example in Eden [BLA85,ALM85] This approach is based on the assumption that the distributed operating system and the various distributed applications running on top of the kernel can make efficient use of the same communication mechanism. According to our opinion based on experiences within the INCAS project this assumption is generally not true.…”