초록 ▼

Apparatus and accompanying methods for optimally using available computer resources, illustratively processing time, and which can be advantageously used for selecting task(s) instances to be precomputed during idle time as well as during other periods of processing activity. Specifically, at an ons

Apparatus and accompanying methods for optimally using available computer resources, illustratively processing time, and which can be advantageously used for selecting task(s) instances to be precomputed during idle time as well as during other periods of processing activity. Specifically, at an onset of each idle-time interval, processing time is allocated to precompute during the remainder of that interval a future task instance, from among a group of such instances then available for precomputation, that will provide the highest fixed or incremental utility. For those task instances which exhibit constant or varying value with time, task selection is based on maximum probability of future occurrence, or net expected value (NEV), respectively, of each such instance. NEV is evaluated as a product of the task instance probability multiplied by a rate of change in the value (EVC flux) to be provided by that task with continued computation time, respectively. This product is assessed, for task instances that exhibit linearly changing value with time, at the onset of each idle-time interval, or, for task instances that exhibit non-linearly changing value, at the onset of each time slice occurring during such an interval. Processing time can be also allocated, at non-idle times, to precomputing a future task instance in lieu of continuing a presently executing task instance, if the future task instance then exhibits a time-discounted NEV that is larger than the EVC flux presently being provided by the currently executing task instance.

대표청구항 ▼

[ I claim:] [1.] In a computer system which exhibits at least one idle time interval during which the system then has processing resources available for processing a task, a method comprising the steps of:defining a list having a plurality of future task instances wherein each of said future task in

[ I claim:] [1.] In a computer system which exhibits at least one idle time interval during which the system then has processing resources available for processing a task, a method comprising the steps of:defining a list having a plurality of future task instances wherein each of said future task instances is expected to yield a result that is likely to be required, subsequent to said one idle time interval, either by a program currently executing on the system or by a current user of the system; andat an onset of said one idle time interval:ranking said instances in the list in terms of their importance so as to define an order among said instances, said order being a function of a likelihood that results of each of said task instances will be needed subsequent to said idle time interval, a cost associated with each of said task instances, or value to be provided subsequent to said idle time interval by each of said task instances, wherein the order dynamically varies for successive idle time intervals in response to specific task instances then constituting the list at the onset of each of said successive idle time intervals and corresponding likelihood of execution, cost or value of each of said specific task instances;selecting, for current execution during said one idle time interval, one of said future task instances, from the list, then having a highest ranking in the order so as to define a selected future task instance; andexecuting said selected future task instance to a fullest extent possible during all remaining time in said one idle time interval and storing a result of the selected future task for use subsequent to said one idle time interval;whereby said selected future task instance is precomputed during said one idle time interval such that, by utilizing the processing resources available during the one idle time interval, run-time delay, that would otherwise and subsequently arise if the selected future task instance needed to be executed subsequent to the idle time interval but was not so precomputed, is reduced.