Policy management for distributed computing and a method for aging statistics

A computer having a CPU and a memory comprising:

• policy program means for causing said CPU to mange the availability to do work of a plurality of cooperating computers that are connected in a network, said policy program means comprising:
• first means for performing a first operation that identifies a set of specific ones of said plurality of cooperating computers as available resources for the performance of work;
• second means for performing a second operation that receives performance related values of said plurality of cooperating computers;
• third means for performing a third operation that derives performance related metrics from said performance related values;
• and
• fourth means for performing a fourth operation that changes said set of specific ones of said plurality of cooperating computers based on said performance related metrics;
• wherein said second operation receives said performance related values over a series of time intervals, and wherein said third operation derives said performance metrics for n periods of each of said intervals, of which the performance metrics of the nth period thereof includes an aggregate of the performance metrics for a current interval plus n−1 of the preceding intervals.

A memory medium for controlling a computer, said memory medium comprising:

• a policy manager program that controls said computer to manage the availability to do work of a plurality of cooperating computers connected in a network, said policy manager program comprising program instructions:
• for controlling said computer to perform a first operation that identifies a set of specific ones of said plurality of cooperating computers as available resources for the performance of work;
• for controlling said computer to perform a second operation that receives performance related values of said plurality of cooperating computers;
• for controlling said computer to perform a third operation that derives performance related metrics from said performance related values;
• and
• for controlling said computer to perform a fourth operation that changes said set of specific ones of said plurality of cooperating computers based on said performance related metrics;
• wherein second operation receives said performance related values over a series of time intervals, and wherein said third operation derives said performance metrics for n periods of each of said intervals, of which the performance metrics of the nth period thereof includes an aggregate of the performance metrics for a current interval plus n−1 of the preceding intervals.

A method comprising:

• managing the availability to do work of a plurality of cooperating computers, which are connected in a network and which comprise a cluster of said cooperating computers connected to a node contained in said network, by:
• (a) identifying a set of specific ones of said plurality of cooperating computers as available resources for the performance of said work;
• (b) receiving performance related values of said plurality of cooperating computers;
• (c) deriving performance related metrics from said performance related values;
• (d) changing said set of specific ones of said plurality of cooperating computers based on said performance related metrics;
• (e) requesting a manager of said cluster to accept additional work or to give up pending work based on said performance related metrics;
• (f) requesting said manager of said cluster to start more work or to run more pieces of an application on one or more of the cooperating computers of said cluster;
• and
• wherein step (d) forms said performance metrics as an aggregation of said values for each of said cooperating computers, wherein step (b) receives said performance related values over a series of time intervals, and wherein step (C) derives said performance metrics for n periods of each of said intervals, of which the performance metrics of the nth period thereof includes an aggregate of the performance metrics for a current interval plus n-−1 of the preceding intervals, wherein step (c) forms said performance metrics as a data structure having n rows that contain the performance metrics of said n periods, respectively, wherein the performance metrics of the nth row of a preceding interval are discarded during a current interval, and wherein said nth row of the preceding interval is used as a first row in the current interval and the remaining ones of said n rows are shifted down one row position.