Throughput
Throughput of a pipeline may be defined as number of results which have been achieved per unit time. It can be referred as:
T = n / [m + (n-1)]. c = E / c
Throughput indicates the computing power of a pipeline.
Utmost efficiency, speed up and throughput are the ideal cases however these are not achieved in practical cases because the speedup is limited because of the subsequent factors:
- Data dependency between successive tasks: There can be dependencies between the instructions of two tasks used in pipeline. E.g. one instruction can't be started till the prior instruction yields the results as both are interdependent. Another Illustration of data dependency would be when that both instructions attempt to modify the same data object. These are termed as data hazards.
- Resource Constraints: When resources aren't accessible at the time of execution then delays are caused in pipelining. E.g. if one common memory is used for both instructions and data and there is need to read and write and fetch an instruction at the same instance then only one can be performed and the other has to wait. Another instance of limited resource such as execution unit that might be busy at the required time.
- Branch Instructions and Interrupts in the program: A program isn't a straight flow of serial instructions. There can be branch instructions which alter the general flow of program, which delays pipelining execution and influences the performance. In the same way, there are interrupts which postpones the execution of subsequent instruction till the interrupt has been serviced. Interrupts and Branches have damaging effects on the pipelining.