The user writes a MOOSE program as a large collection of small tasks that communicate with each other by sending messages through pipes, as shown in Figure 15.3. Each task controls a piece of data, so it can be viewed as an object in the object-oriented sense (hence the name Multitasking Object-oriented OS). The tasks and pipes can be created at any time by any task on any node, so the whole system is completely dynamic.
Figure 15.3: An Executing MOOSE Program Is a Dynamic Network (left) of Tasks
Communicating Through FIFO Buffers Called Pipes (right).
The MOOS II extensions allow one to form groups of tasks called teams that share access to a piece of data. Also, a novel feature of MOOS II is that teams are relocatable, that is, they can be moved from one node to another while they are running. This allows one to perform dynamic load balancing if necessary.
The various subsystems of MOOS II, which together form a complete operating system and programming environment, are shown in Figure 15.4. For convenience, we attempted to preserve a UNIX flavor in the design and were also able to provide support for debugging and performance evaluation because the iPSC/1 hardware has built-in memory protection. Easy interaction with the host is achieved using ICubix, an asynchronous version of Cubix (Section 5.2) that gives each task access to the Unix system calls on the host. The normal C-compilers can be used for programming, and the only extra utility program required is a binder to link the user program to the operating system.
Figure 15.4: Subsystems of the MOOS II Operating System
Despite the increased functionality, the performance of MOOS II on the iPSC/1 turned out to be slightly better than that of Intel's proprietary NX system.