Is there any other motivation for using parallelism? Namely, is there any problem we can solve using parallelism but can't using a single process, given infinite time and finite space(or whatever other measures).
This comment was marked helpful 0 times.
pdixit
One answer is that the ability to solve problems faster due to parallel execution has made some tasks which were previously not possible, possible. People are now attempting to do what they never considered in the past. For example, real-time ray-tracing with massively parallel graphics processors.
If we ignore the above answer, then what you mention is essentially right. Given infinite time, a single processor is capable of doing what a multi-processor can do. But even in a CPU system, the CPU is not the only processing element in the computer. There was always a small micro-controller embedded to handle hard-disk or some logic to handle I/O.
Another motivation for parallelism is power (as mentioned in class). If you finish some work in half the time with 2 core processor, you can shut-down the device. You might think that 2 processors in half the time will consume same power - and you are right. But the 2 processors also share other resources like memory system, hard-disk, display screen of the computer. Shared resources are powered-up if any of the processor is working. So, when task finishes in half the time, you save power by powering down these shared resources. You don't have to do that as a user, unused parts of a systems power down themselves without users knowing about it.
Some systems have parallelism for fault tolerance. They perform the same computation on replicas of same hardware and then compare the results to make sure things were correct. This can be done at fine granularity, such as replicating and ALU or some other sub-unit of a processor core. This is a whole research area.
This comment was marked helpful 3 times.
lazyplus
One more example for the eager need of speedup:
If the weather predicting program needs one day to get the next day's prediction from today's data, then the result would be meaningless. Because once we got the answer, the "next day" was already yesterday.
Is there any other motivation for using parallelism? Namely, is there any problem we can solve using parallelism but can't using a single process, given infinite time and finite space(or whatever other measures).
This comment was marked helpful 0 times.
One answer is that the ability to solve problems faster due to parallel execution has made some tasks which were previously not possible, possible. People are now attempting to do what they never considered in the past. For example, real-time ray-tracing with massively parallel graphics processors.
If we ignore the above answer, then what you mention is essentially right. Given infinite time, a single processor is capable of doing what a multi-processor can do. But even in a CPU system, the CPU is not the only processing element in the computer. There was always a small micro-controller embedded to handle hard-disk or some logic to handle I/O.
Another motivation for parallelism is power (as mentioned in class). If you finish some work in half the time with 2 core processor, you can shut-down the device. You might think that 2 processors in half the time will consume same power - and you are right. But the 2 processors also share other resources like memory system, hard-disk, display screen of the computer. Shared resources are powered-up if any of the processor is working. So, when task finishes in half the time, you save power by powering down these shared resources. You don't have to do that as a user, unused parts of a systems power down themselves without users knowing about it.
Some systems have parallelism for fault tolerance. They perform the same computation on replicas of same hardware and then compare the results to make sure things were correct. This can be done at fine granularity, such as replicating and ALU or some other sub-unit of a processor core. This is a whole research area.
This comment was marked helpful 3 times.
One more example for the eager need of speedup: If the weather predicting program needs one day to get the next day's prediction from today's data, then the result would be meaningless. Because once we got the answer, the "next day" was already yesterday.
This comment was marked helpful 0 times.