Performance tuning a computer system in scaling domains based on quantified scalability

Patent metadata

Field	Value
Publication number	US-12190125-B2
Application number	US-202318105761-A
Country	US
Kind code	B2
Filing date	Feb 3, 2023
Priority date	Feb 3, 2023
Publication date	Jan 7, 2025
Grant date	Jan 7, 2025

Abstract

Systems and methods for performance tuning a computer system in scaling domains based on quantified scalability. A system includes a processor configured to: calculate an estimate of parallel fraction and speedup characteristic in a first domain D1 and in a second domain D2 for an application, the estimate being calculated using system performance measurements generated from previous processing iterations of one or more workloads of the application using a number, n, of cores in the first domain and a remaining number, N−n, of cores in the second domain to obtain performance values X D1 (n) and X D2 (N−n), wherein N represents a total number of cores; calculate the number of cores for the first domain using a quadratic equation generated from the parallel fraction and performance value in each domain; and execute the application in each domain using the number of cores for each domain.

First claim

What is claimed is:

1. A system comprising:
at least one processing device including a processor coupled to a memory;
the at least one processing device being configured to:
calculate an estimate of parallel fraction and speedup characteristic in a first scaling domain D1 and in a second scaling domain D2 for an application executed by a computer system, the estimate being calculated using system performance measurements generated from previous processing iterations of one or more workloads of the application using a number, n, of CPU cores in the first scaling domain and a remaining number, N−n, of CPU cores in the second scaling domain to obtain corresponding performance values X D1 (n) and X D2 (N−n), wherein N represents a total number of CPU cores;
calculate the number of CPU cores for the first scaling domain, the number of CPU cores being calculated using a quadratic equation generated from the parallel fraction and performance value in each scaling domain; and
execute the application in each scaling domain using the number of CPU cores for each scaling domain.

2. The system of claim 1 ,
wherein the performance value for the first scaling domain X D1 (n) is determined based on system performance measurements including the parallel fraction for the first scaling domain and a single core performance value for the first scaling domain X D1 (1), and
wherein the performance value for the second scaling domain X D2 (N−n) is determined based on system performance measurements including the parallel fraction for the second scaling domain and a single core performance value for the second scaling domain X D2 (1).

3. The system of claim 2 , wherein the quadratic equation relates the single core performance value for the first scaling domain with n number of cores allocated, to the single core performance value for the second scaling domain with the remaining number N−n of cores allocated.

4. The system of claim 3 , wherein the quadratic equation yields at most two quadratic solutions, and the number of CPU cores for the first scaling domain is calculated based on the quadratic solution that is between 0 and the total number N of CPU cores.

5. The system of claim 1 , wherein the estimate is calculated using linear regression of values (1/n, 1/X D (n)) generated from trial runs of the application in a given scaling domain.

6. The system of claim 1 , wherein the estimate of parallel fraction and speedup characteristic is calculated based on an expression relating speedup to an inverse of a parallel fraction for the application in a given scaling domain, the parallel fraction being a fraction of the application that can be executed in parallel.

7. The system of claim 6 , wherein the estimate of parallel fraction and speedup characteristic is calculated further based on an inversion of the expression that yields a linear expression for 1/X D (n) in terms of 1/n, the linear expression having slope and intercept values identified by a linear regression and used in substitution calculations to obtain the parallel fraction and a single-core performance value in a given scaling domain.

8. The system of claim 7 , wherein the substitution calculations include (1) calculating the single-core performance value as 1/(m+b), m being the slope value and b being the intercept value, and (2) calculating the parallel fraction as m/(m+b).

9. The system of claim 1 , wherein the parallel fractions in the first and the second scaling domains are estimated using system performance measurements generated from trial runs of the application processing the workloads by selecting a fixed number of CPU cores in a given scaling domain among the first and the second scaling domain and varying the remaining number of CPU cores in another scaling domain among the first and the second scaling domain to obtain the parallel fraction in the other scaling domain.

10. The system of claim 1 , wherein the performance values are values of input/output operations per second (IOPS) as a performance metric for the application.

11. The system of claim 1 , wherein the performance values X D1 (n) and X D2 (N−n) are scaled values calculated by normalizing raw performance values with respect to CPU utilization achieved in the respective scaling domains during respective trials of the application.

12. The system of claim 1 , wherein the CPU cores comprise a plurality of cores in one or more CPUs.

13. The system of claim 1 , wherein the calculated number of CPU cores represents an optimal number of CPU cores for the first scaling domain.

14. A method comprising:
calculating an estimate of parallel fraction and speedup characteristic in a first scaling domain D1 and in a second scaling domain D2 for an application executed by a computer system, the estimate being calculated using system performance measurements generated from previous processing iterations of one or more workloads of the application using a number, n, of CPU cores in the first scaling domain and a remaining number, N−n, of CPU cores in the second scaling domain to obtain corresponding performance values X D1 (n) and X D2 (N−n), wherein N represents a total number of CPU cores;
calculating the number of CPU cores for the first scaling domain, the number of CPU cores being calculated using a quadratic equation generated from the parallel fraction and performance value in each scaling domain; and
executing the application in each scaling domain using the number of CPU cores for each scaling domain.

15. The method of claim 14 ,
wherein the performance value for the first scaling domain X D1 (n) is determined based on system performance measurements including the parallel fraction for the first scaling domain and a single core performance value for the first scaling domain X D1 (1), and
wherein the performance value for the second scaling domain X D2 (N−n) is determined based on system performance measurements including the parallel fraction for the second scaling domain and a single core performance value for the second scaling domain X D2 (1).

16. The method of claim 15 , wherein the quadratic equation relates the single core performance value for the first scaling domain with n number of cores allocated, to the single core performance value for the second scaling domain with the remaining number N−n of cores allocated.

17. The method of claim 16 , wherein the quadratic equation yields at most two quadratic solutions, and the number of CPU cores for the first scaling domain is calculated based on the quadratic solution that is between 0 and the total number N of CPU cores.

18. The method of claim 14 , wherein the parallel fractions in the first and the second scaling domains are estimated using system performance measurements generated from trial runs of the application processing the workloads by selecting a fixed number of CPU cores in a given scaling domain among the first and the second scaling domain and varying the remaining number of CPU cores in another scaling domain among the first and the second scaling domain to obtain the parallel fraction in the other scaling domain.

19. The method of claim 14 , wherein the calculated number of CPU cores represents an optimal number of CPU cores for the first scaling domain.

20. A non-transitory computer-readable storage medium having stored thereon program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform the following steps:
calculating an estimate

Assignees

• Emc Ip Holding Co Llc

Inventors

• Raghavan Rasa
• Morley Steve A

Classifications

• G06F9/44505 Primary

  Configuring for program initiating, e.g. using registry, configuration files · CPC title