Approximate MVA solution system and method for user classes with a known throughput rate

• US 5,345,579 A
• Filed: 10/01/1990
• Issued: 09/06/1994
• Est. Priority Date: 10/01/1990
• Status: Expired due to Term

• Current Assignee: Hewlett-Packard Company (HP Inc.)
• Original Assignee: Hewlett-Packard Company (HP Inc.)
• Inventors: Hynes, Gary
• Primary Examiner(s): Eng, David Y.
• Application Number:
  US07/591,980
• Time in Patent Office: 1,436 Days
• Field of Search: 395/500, 364/148, 364/801, 364/578
• US Class Current: 703/2
• CPC Class Codes: G06F 11/3414 Workload generation, e.g. scripts, playback
• CPC Class Codes: G06F 11/3447 Performance evaluation by modeling
• CPC Class Codes: G06F 2201/865 Monitoring of software

Abstract

The present invention is an application computer software program which operates on a computer software platform. The application computer software program of the present invention comprises systems and methods for solving models of computer systems. In other words, the application computer software program of the present invention comprises an implementation of a model solution. The present invention solves the problems inherent in prior solutions by advantageously using fixed class workloads, rather than transaction workloads, to represent those terminal and batch workloads which are specified with throughputs, rather than population. Additionally, the present invention advantageously implements a fixed class algorithm as the model solution.

Claims

1. A computer-based, analytic modeling system for predicting the manner in which a computer system will operate under future conditions and loads, said computer-based modeling system comprising:

(1) first means for entering information, wherein said information represents an analytic model of the computer system, and wherein said model comprises fixed class workloads and non-fixed class workloads;

(2) second means for solving said model to generate performance metrics, said performance metrics representing indicia for predicting the manner in which the computer system will operate under future conditions and loads; and

(3) third means for displaying said information and said performance metrics;

said second means comprising

fourth means for setting first queue lengths to initial values representing estimates of average queue lengths at service stations as seen by outside observers, the service stations representing elements in the computer system,

fifth means for initializing populations of said fixed class workloads to predetermined values,

sixth means for setting second queue lengths to estimates of queue lengths at the service stations as seen by arriving customers,

seventh means for calculating response times for said fixed and non-fixed class workloads using said second queue lengths,

eighth means for summing said response times for said fixed and non-fixed class workloads,

ninth means for calculating new population values for said fixed class workloads using said response times sums for said fixed class workloads,

tenth means for calculating throughputs for said non-fixed class workloads using said response times sums for said non-fixed class workloads,

eleventh means for calculating new values for said first queue lengths using said response times sums for said fixed and non-fixed class workloads, and

twelfth means for testing for convergence of said first queue lengths by comparing said new first queue length values with said initial first queue length values, said performance metrics being accurate measures of how the computer system will operate under future conditions and loads once said convergence occurs.

2. A computer-based, analytic model solution system for solving a model of a computer system such that operation of the computer system under future loads and conditions can be predicted, wherein said model comprises fixed class workloads and non-fixed class workloads, said computer-based model solution system comprising:

first means for setting first queue lengths to initial values representing estimates of average queue lengths at service stations as seen by outside observers, the service stations representing elements in the computer system;

second means for initializing populations of said fixed class workloads to predetermined values, wherein said second means comprises means for initializing batch fixed class workload populations to a first initial value and terminal fixed class workload populations to a second initial value;

third means for setting second queue lengths to estimates of queue lengths at the service stations as seen by arriving customers;

fourth means for calculating response times for said fixed and non-fixed class workloads using said second queue lengths;

fifth means for summing said response times for said fixed and non-fixed class workloads;

sixth means for calculating new population values for said fixed class workloads using said response times sums for said fixed class workloads;

seventh means for calculating throughputs for said non-fixed class workloads using said response times sums for said non-fixed class workloads;

eighth means for calculating new values for said first queue lengths using said response times sums for said fixed and non-fixed class workloads; and

ninth means for testing for convergence of said first queue lengths by comparing said new first queue length values with said initial first queue length values, said performance metrics being accurate measures of how the computer system will operate under future conditions and loads once said convergence occurs.

3. The computer-based model solution system of claim 2, wherein said eighth means comprises:

means for comparing said new values to an upper bound; and

means for setting said new values to said upper bound if said new values exceed said upper bound.

4. A computer-based method for solving an analytic model of a computer system such that operation of the computer system under future loads and conditions can be predicted, wherein said model comprises fixed class workloads and non-fixed class workloads, said method executed using a computer and comprising the steps of:

(a) setting first queue lengths to initial values representing estimates of average queue lengths at service stations as seen by outside observers, the service stations representing elements in the computer system;

(b) initializing populations of said fixed class workloads to predetermined values;

(c) setting second queue lengths to estimates of queue lengths at the service stations as seen by arriving customers;

(d) calculating response times for said fixed and non-fixed class workloads using said second queue lengths;

(e) summing said response times for said fixed and non-fixed class workloads;

(f) calculating new population values for said fixed class workloads using said response times sums for said fixed class workloads;

(g) if said new fixed class workload population values exceed an upper bound, then setting said new fixed class workload population values to said upper bound;

(h) calculating throughputs for said non-fixed class workloads using said response times sums for said non-fixed class workloads;

(i) calculating new values for said first queue length using said response times sums for said fixed and non-fixed class workloads; and

(j) testing for convergence of said first queue lengths by comparing said new first queue length values with said initial first queue length values, said performance metrics being accurate measures of how the computer system will operate under future conditions and loads once said convergence occurs.

5. The method of claim 4, further comprising the step of calculating throughputs for said fixed class workloads.

6. A computer-based method for solving a model of a computer system such that operation of the computer system under future loads and conditions can be predicted, wherein said model comprises fixed class workloads and non-fixed class workloads, said method executed using a computer and comprising the steps of:

(a) initializing parameters, comprising the steps of;

(1) setting first queue lengths to initial values representing estimates of average queue lengths at service stations as seen by outside observers, the service stations representing elements in the computer system;

(2) initializing populations of said fixed class workloads to predetermined values; and

(3) indicating whether said method is operating in a regular mode or a hypermode;

(b) setting second queue lengths to estimates of queue lengths at the service stations as seen by arriving customers;

(c) if said method is operating in said regular mode, then calculating performance metrics for all of said fixed and non-fixed class workloads, step (c) comprising the steps of;

(i) calculating response times for all of said fixed and non-fixed class workloads using said second queue lengths;

(ii) summing said response times for all of said fixed and non-fixed class workloads;

(iii) calculating new population values for all of said fixed class workloads using said response times sums for said fixed class workloads;

(iv) calculating throughputs for all of said non-fixed class workloads using said response times sums for said non-fixed class workloads;

(d) if said method is operating in said hypermode, then calculating performance metrics for any fixed and non-fixed class workloads which have not converged, step (d) comprising the steps of;

(i) calculating response times for any fixed and non-fixed class workloads which have not converged using said second queue lengths;

(ii) summing said response times for any fixed and non-fixed class workloads which have not converged;

(iii) calculating new population values for any fixed class workloads which have not converged using said response times sums for said fixed class workloads;

(iv) calculating throughputs for any non-fixed class workloads which have not converged using said response times sums for said non-fixed class workloads;

(e) calculating new values for said first queue length using said response times sums for said fixed and non-fixed class workloads; and

(f) testing for convergence of said first queue lengths by comparing said new first queue length values with said initial first queue length values, said performance metrics being accurate measures of how the computer system will operate under future conditions and loads once said convergence occurs.

7. The method of claim 6, wherein step (a)(3) comprises the step of setting a mode flag to indicate that said method is operating in said regular mode.

8. The method of claim 7, wherein step (a) further comprises the steps of:

(4) initializing an iteration counter; and

(5) initializing a convergence array, such that said convergence array indicates that said fixed and non-fixed class workloads have not converged;

and wherein step (f) comprises the steps of;

(1) identifying which of said fixed and non-fixed class workloads have converged, and indicating in said convergence array said identified fixed and non-fixed class workloads that have converged;

(2) incrementing said iteration counter; and

(3) setting said mode flag to said hypermode when said iteration counter exceeds a predetermined value.

9. A computer-based method for solving a model of a computer system such that operation of the computer system under future loads and conditions can be predicted, wherein said model comprises fixed class workloads and non-fixed class workloads, said method executed using a computer and comprising the steps of:

(a) setting first queue lengths to initial values representing estimates of average queue lengths at service stations as seen by outside observers, the service stations representing elements in the computer system;

(b) initializing populations of said fixed class workloads to predetermined values;

(c) setting second queue lengths to estimates of queue lengths at the service stations as seen by arriving customers;

(d) calculating response times for said fixed and non-fixed class workloads using said second queue lengths;

(e) summing said response times for said fixed and non-fixed class workloads;

(f) selecting a current fixed class workload from said fixed class workloads;

(g) calculating a new population value for said current fixed class workload using said response times sums for said fixed class workloads;

(h) using said new population value and a previous population value of said current fixed class workload to calculate a population change of said current fixed class workload;

(i) determining a convergence direction of said current fixed class workload based on said population change;

if said convergence direction oscillates, then proceeding to step (n);

(k) calculating a current population value for said current fixed class workload based on said population change;

(l) calculating an inflated population for said current fixed class workload using said current population value;

(m) distributing said inflated population;

(n) performing steps (f)-(m) for each of said fixed class workloads;

(o) calculating throughputs for said non-fixed class workloads using said response times sums for said non-fixed class workloads;

(p) calculating new values for said first queue lengths using said response times sums for said fixed and non-fixed class workloads; and

(q) testing for convergence of said first queue lengths by comparing said new first queue length values with said initial first queue length values, said performance metrics being accurate measures of how the computer system will operate under future conditions and loads once said convergence occurs.

10. A computer-based method for solving a model of a computer system such that operation of the computer system under future loads and conditions can be predicted, wherein said model comprises fixed class workloads and non-fixed class workloads, said method executed using a computer and comprising the steps of:

(a) setting first queue lengths to initial values representing estimates of average queue lengths at service stations as seen by outside observers, the service stations representing elements in the computer system;

(b) initializing populations of said fixed class workloads to predetermined values;

(c) creating composite classes, wherein said composite classes comprise subgroups of said fixed class workloads;

(d) setting second queue lengths to estimates of queue lengths at the service stations as seen by arriving customers;

(e) calculating response times for said fixed and non-fixed class workloads using said second queue lengths;

(f) summing said response times for said fixed and non-fixed class workloads;

(g) selecting a current composite class from said composite classes;

(h) calculating new population values for said fixed class workloads in said current composite class using said response times sums for said fixed class workload;

(i) calculating population changes for said fixed class workloads in said current composite class using said new population values and previous population values for said fixed class workloads in said current composite class;

calculating a net population change for said current composite class using said population changes;

(k) determining a convergence direction based on said net population change;

(l) if said convergence direction oscillates, then proceeding to step (p);

(m) calculating current population values for said fixed class workloads in said current composite class based on said population changes;

(n) calculating inflated populations for said fixed class workloads in said current composite class based on said current population values;

(o) distributing said inflated populations;

(p) performing steps (g)-(o) for each of said composite classes;

(q) calculating throughputs for said non-fixed class workloads using said response times sums for said non-fixed class workloads;

(r) calculating new values for said first queue lengths using said response times sums for said fixed and non-fixed class workloads; and

(s) testing for convergence of said first queue lengths by comparing said new first queue length values with said initial first queue length values, said performance metrics being accurate measures of how the computer system will operate under future conditions and loads once said convergence occurs.

11. The method of claim 10, wherein said step (c) comprises the step of assigning said fixed class workloads with an identical priority level to one of said composite classes.

12. The method of claim 10, wherein said step (c) comprises the step of assigning said fixed class workloads operating in a FCFS system to one of said composite classes.