Accuracy in a prediction of resource usage of an application in a virtual environment

• US 8,131,519 B2
• Filed: 09/30/2008
• Issued: 03/06/2012
• Est. Priority Date: 09/30/2008
• Status: Active Grant

• Current Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Wood, Timothy W., Cherkasova, Ludmila
• Primary Examiner(s): Thangavelu, Kandasamy
• Application Number:
  US12/241,913
• Publication Number:
  US 20100082320A1
• Time in Patent Office: 1,253 Days
• Field of Search: 703/2, 703/13, 718/102, 718/1, 718/104, 711/170, 711/114, 707/206, 713/322
• US Class Current: 703/2
• CPC Class Codes: G06F 11/3428 Benchmarking
• CPC Class Codes: G06F 11/3457 Performance evaluation by simulation
• CPC Class Codes: G06F 2201/815 Virtual middleware or OS functionality using virtual machines to implement generic software techniques for error detection or fault masking G06F11/1484
• CPC Class Codes: G06F 2209/5019 Workload prediction
• CPC Class Codes: G06F 2209/508 Monitor
• CPC Class Codes: G06F 9/5011 the resources being hardware resources other than CPUs, Servers and Terminals

Abstract

Described herein is a system for improving accuracy in a prediction of resource usage of an application running in a virtual environment. The system comprises a first hardware platform implementing a native hardware system in which an application natively resides and executes, the native hardware system operating to execute a predetermined set of benchmarks that includes at least one of a network-intensive workload and a disk-intensive workload, a second hardware platform implementing a virtual environment therein, the virtual environment operating to execute the predetermined set of benchmarks, and a third hardware platform operating to collect first resource usage traces from the first hardware platform that result from the execution of the predetermined set of benchmarks in the native hardware system and second resource usage traces from the second hardware platform that result from the execution of the predetermined set of benchmarks in the virtual environment. The third hardware platform further operating to perform a linear regression computation to generate at least one prediction model that predicts a resource usage of the application executing in the virtual environment based on the collected first and second resource usage traces.

Claims

1. A method for improving accuracy in a prediction of resource usage of an application running in a virtual environment, comprising:

providing a predetermined set of benchmarks, wherein the predetermined set of benchmarks includes at least one of a network-intensive workload and a disk-intensive workload;

executing the predetermined set of benchmarks in a native hardware system in which the application natively resides;

executing the predetermined set of benchmarks in the virtual environment;

collecting first traces of first resource utilization metrics in the native hardware system based on the execution of the predetermined set of benchmarks in the native hardware system;

collecting second traces of second resource utilization metrics in the virtual environment based on the execution of the predetermined set of benchmarks in the virtual environment;

performing a linear regression computation to generate at least one prediction model that maps the first traces of resource utilization metrics to the second traces of resource utilization metrics; and

predicting a resource usage of the application running in the virtual environment with the prediction model.

2. The method of claim 1, wherein performing the linear regression computation to generate the at least one prediction model comprises:

performing a robust linear regression computation to generate the at least one prediction model.

3. The method of claim 2, wherein performing the robust linear regression computation to generate the at least one prediction model comprises:

performing an iteratively reweighted least squares regression computation to generate the at least one prediction model.

4. The method of claim 3, wherein performing the iteratively reweighted least squares regression computation to generate the at least one prediction model comprises:

performing the iteratively reweighted least squares regression computation using a bisquare weighting function to generate the at least one prediction model.

5. The method of claim 1, wherein performing the linear regression computation to generate the at least one prediction model comprises;

performing a stepwise linear regression computation to generate the at least one prediction model.

6. The method of claim 5, wherein performing the stepwise linear regression computation to generate the at least one prediction model comprises:

a) selecting one of the first resource utilization metrics;

b) performing a statistical test on the at least one prediction model with the selected resource utilization metric included therein;

c) determining whether the statistical test passes or fails;

d) upon the passing of the statistical test, including the selected one of the first resource utilization metrics and

(e) iteratively repeating steps a) through d) for remaining ones of the first resource utilization metrics.

7. The method of claim 6, wherein selecting one of the first resource utilization metrics comprises:

performing a least squares regression computation to generate a prediction model associated with each of the first resource utilization metrics;

selecting the one metric from the first resource utilization metrics that has the associated prediction model with the lowest mean squared error.

8. The method of claim 6, further comprising:

upon the failure of the statistical test, excluding the selected one of the first resource utilization metrics and ceasing the step of performing the stepwise linear regression computation.

9. The method of claim 1, further comprising:

hosting the virtual environment on a hardware platform that has different hardware configurations from the native hardware platform.

10. A system for improving accuracy in a prediction of resource usage of an application running in a virtual environment, comprising:

a first hardware platform implementing a native hardware system in which an application natively resides and executes, the native hardware system operating to execute a predetermined set of benchmarks that includes at least one of a network-intensive workload and a disk-intensive workload;

a second hardware platform implementing a virtual environment therein, the virtual environment operating to execute the predetermined set of benchmarks;

a third hardware platform operating to collect first resource usage traces from the first hardware platform that result from the execution of the predetermined set of benchmarks in the native hardware system and second resource usage traces from the second hardware platform that result from the execution of the predetermined set of benchmarks in the virtual environment;

wherein the third hardware platform further operating to perform a linear regression computation to generate at least one prediction model that predicts a resource usage of the application executing in the virtual environment based on the collected first and second resource usage traces.

11. The system of claim 10, wherein the linear regression computation is an iteratively reweighted least squares regression computation that uses a bisquare weighting function to generate the at least one prediction model.

12. The system of claim 10, wherein the linear regression computation is a stepwise linear regression computation that uses a robust linear regression computation therein.

13. The system of claim 10, wherein the predetermined set of benchmarks further include a computation intensive workload.

14. The system of claim 13, wherein the first resource usage traces comprises at least one of:

an aggregate of a percentage of processor utilization in the native hardware system that is attributed to a utilization of hardware processing resources in the first hardware platform by a user space in the native hardware system, a percentage of processor utilization in the native hardware system that is attributed to a utilization of processing resources in the first hardware platform by a kernel in the native hardware system, and a percentage of processor utilization in the native hardware system that is attributed to a utilization of processing resources in the first hardware platform by an input/output (I/O) wait therein;

an aggregate of a network packet rate of data packets received from a data network by the first hardware platform, a network packet rate of data packets transmitted through the data network by the first hardware platform, a network byte rate of data bytes received from the data network by the first hardware platform, and a network byte rate of data bytes transmitted through the data network by the first hardware platform; and

an aggregate of a rate of disk read requests that is received by the first hardware platform for accessing a disk memory therein, a rate of disk write requests that is received by the first hardware platform for accessing a disk memory therein, a rate of memory blocks requested for reading that is received by the first hardware platform for accessing a disk memory therein, and a rate of memory blocks requested for writing that is received by the first hardware platform for accessing a disk memory therein.

15. A non-transitory computer readable medium storing computer executable programming code which when executed by a computer processor improves accuracy in a prediction of resource usage of an application running in a virtual environment, the programming code comprising:

programming code for providing a predetermined set of benchmarks, wherein the predetermined set of benchmarks includes at least one of a network-intensive workload and a disk-intensive workload;

programming code for executing the predetermined set of benchmarks in a native hardware system in which the application natively resides;

programming code for executing the predetermined set of benchmarks in the virtual environment;

programming code for collecting first traces of first resource utilization metrics in the native hardware system based on the execution of the predetermined set of benchmarks in the native hardware system;

programming code for collecting second traces of second resource utilization metrics in the virtual environment based on the execution of the predetermined set of benchmarks in the virtual environment;

programming code for performing a linear regression computation to generate at least one prediction model that maps the first traces of resource utilization metrics to the second traces of resource utilization metrics; and

programming code for predicting a resource usage of the application running in the virtual environment with the prediction model.