Algorithms and Parallel Computing

3.12 SUMMARY OF PARALLEL ARCHITECTURES

The previous sections briefly explained five parallel processor systems that are widely used:

• Shared memory
• Distributed memory
• SIMD
• Systolic
• Multicore

It is hard to uniquely classify each type; for example, SIMD could be built on top of a shared memory system. We can summarize the salient features of these multiprocessors in the following points:

1. All multiprocessors, except systolic processors, communicate using an interconnection network that can be easily identified.

2. Systolic processors have neighbor-to-neighbor connections and few global buses.

3. All multiprocessors, except systolic processors, are more general purpose in nature compared with SIMD. They implement all sorts of tasks and algorithms.

4. Systolic processors are designed to execute a specific algorithm. The algorithm dictates several details, such as the type of interprocessor communication, the I/O data timing, and the feeding or extraction points of the I/O data.

5. A multicore system uses accelerator cores to implement special tasks that need to be implemented at a high rate. For example, we could have a GPU in a multicore system to implement intensive graphic processing tasks. Such accelerator cores are built using systolic processors.

3.13 PROBLEMS

3.1. What is the main communication mechanism between processors in a shared memory multiprocessor system?

3.2. What are the main issues in shared memory processor systems?

3.3. What is the main disadvantage of shared ...