Types of Parallel Systems
Shared memory systems
- Multiple processors, single memory space
- Easier to program for parallelism:not much harder than traditional serial machines such as PCs, workstations
- Many shared memory machines (including all at SDSC, such as Cray T90, Sun E10000, or Tera MTA ) are symmetric multiprocessors (SMPs) , meaning they have uniform memory access.
Distributed memory systems
- Memory is distributed with processors
- More difficult to program, but more scalable to larger numbers of processors (if code doesn’t require much communication)
- Exemplified by massively parallel processors (MPPs) such as SDSC Cray T3E and SDSC IBM SP, and also clusters