- Parallel processing:
- Parallel processing = executing multiple instructions to perform a common task currently
- Architectural requirement for
parallel processing:
- Multiple CPUs
- Memory that can provide (transfer) instructions to the multiple CPUs (so they can execute simultaneously)
- Multiple CPUs
- Categories of
parallel computing
architectures:
- Shared memory
parallel processors:
- All processors (CPUs) use (= share) the same memory
- All processors (CPUs) use (= share) the same disks
Graphically:
Note:
- Shared memory processors
are very expensive to
build:
- You need a high bandwidth switching network to connect the (shared) memory to all the processoes
- Shared disk
parallel processors:
- Each processor (CPUs) uses its own memory
- All processors (CPUs) use (= share) the same disks
Graphically:
- Shared nothing
parallel processors:
- Each processor (CPUs) uses its own memory
- Each processor (CPUs) uses its own disk drives
Graphically:
Note:
- Shared nothing processors
are very cheap to
build:
- You can use a Ethernet router as switching network to connect the (stand-alone) computers
- Shared memory
parallel processors:
- The architecture
of choice for
parallel
data processing is:
- Shared nothing processors:
- Shared nothing processors: