IO bound
|
|
IO bound refers to a condition in which the time it takes to complete a computation is determined principally by the speed of input/output(IO) operations to memory. Compare to Compute bound. This circumstance arises whenever the amount of data necessary to the computation (or generated by the computation) is much larger than what can be held in main memory at one time. This is a common circumstance, and the majority of computer programs are IO bound.
IO, sometimes written as I/O, refers to operations that transfer data into or out of the main memory of a computer from or to peripheral devices. Peripheral devices include, for instance, magnetic storage media, typically a disc drive.
Compared to main memory, data transfers to peripherals usually involve much greater latency, the time between the request for information and when it is available. Long latency is mostly caused by the time to get the transfer started. It is also affected by the fact that peripherals sometimes move data at a slower rate than the rate that can be achieved between main memory and the computer itself. Peripheral transfers also usually move larger blocks of memory at a time than transfers between main memory and the computer. The time to complete a large transfer also increases latency. In some circumstances, techniques such as buffering or spooling or multi-tasking are used to partially compensate for the longer latency of peripheral transfers.
It is possible for a computer program to alternate between being IO bound and being compute bound during different phases of the job, but this usually means that the overall computation is IO bound. One goal of multitasking is to allow processor utilization to be high by having a few compute bound jobs to work on while IO bound jobs are stalled. The goal of high processor utilization was more urgent when processors cost millions than when processors cost hundreds of dollars.