next up previous contents index
Next: 15.2.1 Design of MOOSE Up: High-Level Asynchronous Software Previous: 15.1 Asynchronous Software Paradigms

MOOS II: An Operating System forDynamic Load Balancing on the iPSC/1


Applications involving irregular time behavior or dynamically varying data structures are difficult to program using the crystalline model or its variants. Examples are dynamically adaptive grids for studying shock waves in fluid dynamics, N-body  simulations of gravitating systems, and artificial intelligence applications, such as chess. The few applications in this class that have been written typically use custom designed operating systems and special techniques.

To support applications in this class, we developed a new, general-purpose operating system called MOOSE  for the Mark II hypercube [Salmon:88a], and later wrote an extended version called MOOS II for the Intel iPSC/1 [Koller:88b]. While the MOOSE system was fairly convenient for some applications, it became available at a time when the Mark II and iPSC/1 were falling into disuse because of uncompetitive performance. The iPSC/1 was used for MOOS II for two reasons: It had the necessary hardware support on the node, and, because of low performance, it had little production use for scientific simulation. Thus, we could afford to devote the iPSC/1 to the ``messy'' process of developing a new operating system which rendered the machine unusable to others for long periods of time. Only one major application was ever written using MOOSE (Ray tracing,  [Goldsmith:88a], mentioned in Section 14.1 [Salmon:88c]), and only toy applications were written using MOOS II. Its main value was therefore as an experiment in operating system design and some of its features are now incorporated in Express (Section 5.2). The lightweight threads pioneered in MOOSE are central to essentially all new distributed- and shared-memory computing models-in particular MOVIE, described in Chapter 17.

Guy Robinson
Wed Mar 1 10:19:35 EST 1995