System for improving the performance of a computer software application in a server network

• US 20080270526A1
• Filed: 04/24/2007
• Published: 10/30/2008
• Est. Priority Date: 04/24/2007
• Status: Active Grant

• Current Assignee: HyPerformix, Inc. (Broadcom, Inc.)
• Original Assignee: HyPerformix, Inc. (Broadcom, Inc.)
• Inventors: Reynolds, James, Barnett, Paul T.
• Granted Patent: US 7,680,916 B2
• Time in Patent Office: Days
• US Class Current: 709/203
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46

Abstract

Method and apparatus for separating background utilization from transaction resource consumption to simulate resource costs on system servers including a computer application server is taught. Transaction throughput combined with resource consumption data. Regression analysis on the resulting data set provides information useful for determining performance and for creating discrete event simulations. In a first embodiment, data from an application server is combined with the application transaction log to find a background utilization on the application servers and to simulate transaction resource cost with a background load. In a second embodiment, data from servers other than the application server are combined with the application transaction log to estimate the resource cost per transaction for a given resource in the system. In a third embodiment, data from servers other than the application server are combined with the application transaction log to simulate resource cost for a given resource based on transaction throughput.

Claims

1. A system for creating a set of predicted resource utilizations of a computer application wherein the computer application operates as a software program on a system of servers configured to operate a set of software programs;

the set of resources containing a set of local resources and a set of remote services;

wherein the set of local resources is connected to the set of application servers;

wherein the set of remote resources is connected to the set of networked servers;

the system of servers include a set of application servers which are connected to a set of networked servers by a communications network, and wherein the computer application utilizes a set of system resources the set of system resources contain a set of local resources and a set of remote resources;

wherein the set of local resources is connected to the set of application servers;

wherein the set of remote resources is connected to the set of networked servers;

the set of local resources containing a local CPU resource for executing software programs, a local RAM resource for storing software programs and for storing dynamic data, a local data storage resource for storing persistent data, and a local data pipe resource for communicating with the set of software programs by a set of transactions;

the set of remote resources containing a remote CPU resource for executing software programs, a remote RAM resource for storing software programs and for storing dynamic data, a remote data storage resource for storing persistent data, and a remote data pipe resource for communicating with at least one networked server in the set of networked servers and at least one application server in the set of application servers;

each of the set of transactions including an exchange of data associated with the computer application;

wherein at least one of the set of networked servers services at least one transaction of the set of transactions and services a set of requests from at least one network server in the set of networked servers;

the set of transactions causing usage of the set of system resources;

wherein a transaction log associated with the computer application is generated and contains a list of transaction information measured by the computer application;

the list of transaction information describing resource usage by the computer application;

wherein a set of system resource monitors resides on the system of servers, and each system resource monitor is associated with a particular server in the system of servers, each system resource monitor operating on at least one software program in the set of software programs; and

wherein at least one system resource monitor of the set of system resource monitors measures a set of resource utilization data for the particular server of the set of servers with which the system resource monitor is associated, the system comprising;

a. a performance criteria describing an expected resource utilization by the computer application;

b. a system resource utilization data repository for storing and organizing the set of resource utilization data, the system resource utilization data repository residing on and operating on at least one server in the system of servers;

c. a system model generator residing on and operating on at least one server in the system of servers, the system model generator creating a set of predicted resource utilizations for the system of servers, derived from the set of resource utilization data and from the transaction log.

2. The system of claim 1 wherein the set of system resource utilization data contains a set of measured resource utilizations, ρ

(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for the system resources i in the set of local resources and wherein any given local resource in the set of local resources is comprised of a plurality of n(i) entities;

wherein the transaction log contains a set of measured resource costs R_j(t) corresponding to a group of transactions of the set of transactions occurring at a plurality of times t and for the system resources j in the set of local resources, the system model generator comprising a computer software program residing on at least one server in the set of servers and programmed to;

a. compute a resource cost, C(i, t′

), for each system resources i in the set of system resources according to the formula,

C(i,t′

)=ρ

(i,t′

)·

Δ

t′

·

n_i

b. compute a total system resource cost, C_sys(i, T), for each system resource i in the set of local resources, for a test period T, by summing the resource cost for the test period T according to the formula,

3. The system of claim 2 wherein the system model generator is further programmed to submit the set of constant background load values to a discrete event simulator.

4. The system of claim 1 wherein the set of system resource utilization data contains a set of measured resource utilizations, ρ

(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for the system resources i in the set of local resources and wherein any given local resource in the set of local resources is comprised of a plurality of n(i) entities;

wherein the transaction log contains a set of measured resource costs R_j(t) corresponding to a group of transactions of the set of transactions occurring at a plurality of times t and for the system resources j in the set of local resources, the system model generator comprising a computer software program residing on at least one server in the set of servers and programmed to;

a. compute a resource cost, C(i,t′

), for each system resources i in the set of systems resources according to the formula

C(i,t′

)=ρ

(i,t′

)·

Δ

t′

·

n(i)

b. compute a total system resource cost, C_sys(i, T), for each system resource i in the set of local resources, for a given test period T, by summing the resource cost for the test period T according to the formula,

5. The system of claim 1 wherein the set of system resource utilization data contains a set of measured resource utilizations, ρ

_s(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for system resources i of the set of remote resources and wherein any given remote resource is comprised of a plurality of n_s(i) entities;

wherein the transaction log contains a set of measured resource costs R_j(t) corresponding to a group of transactions of the set of transactions occurring at a plurality of times t and for a group of system resources j contained in the set of remote resources, the system model generator comprising a computer software program residing on at least one server in the set of servers and programmed to;

a. compute the resource cost C_s(i, t′

) for each server s in the set of networked servers, for all system resources i associated with the set of networked servers, using the system resource utilization data and the formula,

C_s(i,t′

)=ρ

_s(i,t′

)·

Δ

t′

·

n_s(i);

b. compute total server resource cost C_s(i, T) for all system resource i in the set of system resources on each server s in the set of networked servers for a given test period T by summing the resource costs occurring within the test period T according to the formula,

6. The system of claim 5 wherein the system model generator is programmed to submit the average transaction resource cost as R_s(i) to the discrete event simulator.

7. The system of claim 1 wherein the set of system resources contains a set of measured resource utilizations, ρ

_s(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for system resources i of the set of remote resources and wherein a given remote resource is comprised of a plurality of n_s(i) entities;

wherein the transaction log contains a set of measured resource costs R_j(t) corresponding to a group of transactions occurring at a plurality of times t from the set of transactions; and

for a group of system resources j contained in the set of remote resources, a system model generator comprising a computer software program residing on at least one server in the set of servers and programmed to;

a. compute the resource cost Cs (i, t′

) for each server s in the set of networked servers, for all system resources i associated with the set of networked servers, using the system resource utilization data and the formula,

C_s(i,t′

)=ρ

_s(i,t′

)·

Δ

t′

·

n_s(i);

b. distribute the set of transactions from the transaction log into a set of τ

(t′

) transactions occurring in the time intervals Δ

t′

,

c. perform a linear regression to fit a straight line function to the resource costs C_s(i,t′

) for each server s as a function of τ

(t′

), for each system resource i of the set of system resources according to the formula,

C_s(i,t′

)=m(i,s)·

τ

(t′

)+b(i,s);

where the m(i, s) is the computed slope from the linear regression analysis and b(i, s) is the computed intercept from the linear regression analysis and where a set of correlation coefficients r²(i, s) are generated for each linear regression,

d. compare the set of correlation coefficients r²(i,s) to a predetermined threshold; and

,

e. calculate a resource cost for the given resource i and server s.

8. The system of claim 7 wherein the system model generator is additionally programmed to calculate resource costs according to the formula, C(i,s)=R(i,s)*τ

+C₀(i,s), for each system resource i of the set of system resources and each server s of the system of servers, for a set of transaction throughput values, τ

, and wherein R(I,s)=M(I,s) and C_O(i,s)=b(i,s).

9. The system of claim 7 wherein the system model generator is additionally programmed to calculate background resource utilization according to ρ

₀(i,s)=C₀(i,s)/T, for test period T, for all resources i and for all servers s and where C_O(i,s)=b(i,s).

10. The system of claim 7 wherein the system model generator is additionally programmed to simulate transaction resource cost as R(i, s) for each system resource i of the set of system resources and each server s of the system of servers, and where R(i,s)=M(i,s).

11. The system of claim 7 wherein the system model generator is further programmed to submit the resource cost to a discrete event simulator.

12. The system of claim 1 wherein the set of system resource utilization data contains a set of measured resource utilizations, ρ

(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for the system resources i in the set of local resources and wherein any given local resource in the set of local resources is comprised of a plurality of n(i) entities;

wherein the transaction log contains a set of measured resource costs R_j(t) corresponding to a group of transactions of the set of transactions occurring at a plurality of times t and for the system resources j in the set of local resources, the system model generator comprising a computer software program residing on at least one server in the set of servers and programmed to;

a. compute a resource cost, C(i,t′

), for each system resources i in the set of systems resources according to the formula

C_sys(i,t′

)=ρ

(i,t′

)·

Δ

t′

·

n(i)

b. compute the total application resource cost, Capp (j, T), during the test period T using the formula,

13. A method for improving the performance of a computer application wherein the computer application operates as a software program on a first configuration of a system of servers operating on a set of software programs;

the system of servers including a set of application servers which are connected to a set of networked servers by a communications network, and wherein the computer application utilizes a set of system resources;

the set of system resources including a set of local resources and a set of remote resources;

the set of local resources connected to the set of application servers, and wherein the computer application utilizes a set of remote resources connected to the set of networked servers;

the set of local resources containing a local CPU resource for executing software programs, a local RAM resource for storing software programs and for storing dynamic data, a local data storage resource for storing persistent data, and a local data pipe resource for communicating with the set of software programs by a set of transactions;

each of the set of transactions including an exchange of data associated with the computer application;

wherein at least one of the set of networked servers services at least one transaction of the set of transactions and services a set of requests from at least one network server of the set of networked servers;

the set of transactions causing the usage of the set of system resources and the set of remote resources;

wherein the set of remote resources contains a remote CPU resource for executing software programs, a remote RAM resource for storing software programs and for storing dynamic data, a remote data storage resource for storing persistent data, and a remote data pipe resource for communicating with at least one networked server in the set of networked servers and at least one application server in the set of application servers; and

wherein a transaction log associated with the computer application is generated and contains a list of transaction information measured by the computer application;

the list of transaction information describing resource usage by the computer application;

wherein a set of system resource monitors resides on the system of servers, each system resource monitor is associated with a particular server in the system of servers, each system resource monitor operating on at least one software program in the set of software programs; and

wherein at least one system resource monitor of the set of system resource monitors measures a set of resource utilization data for the particular server of the set of servers with which the system resource monitor is associated, the method comprising;

a. causing the computer application to run under a set of load conditions;

b. collecting a transaction log from the computer application;

c. collecting the set of resource utilization data and storing the resource utilization data in a system resource utilization repository;

d. generating a system model by combining the list of transaction information with the resource utilization data;

e. running a discrete event simulation based on the system model to create a predicted system resource usage by the computer application;

f. comparing the predicted system resource usage to a desired performance criteria for system resource usage; and

,

g. changing the system of servers to a second configuration which meets the desired performance criteria for system resource usage.

14. The method of claim 13 wherein the set of system resource utilization data contains a set of measured resource utilizations, ρ

(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for the system resources i in the set of local resources and wherein any given local resource in the set of local resources is comprised of a plurality of n(i) entities;

wherein the transaction log contains a set of measured resource costs, R(i, t), corresponding to a group of transactions of the set of transactions occurring at a plurality of times t and for the system resources i in the set of local resources, the step of generating a system model further comprising;

a. computing from the set of system resource utilization data, a resource cost, C(i, t′

), for each system resources i in the set of system resources according to the formula,

C(i,t′

)=ρ

(i,t′

)·

Δ

t′

·

n_i

b. computing a total system resource cost, C_sys(i, T), for each system resource i in the set of local resources, for a test period T, by summing the resource cost for the test period T according to the formula,

15. The method of claim 14 further comprising submitting the background load for each system resource i of the set of system resources, during a simulated transaction, as a random process using as a mean the set of constant background load values ρ

₀(i) to a discrete event simulator.

16. The method of claim 13 wherein the set of system resource utilization data is comprised of a set of measured resource utilizations, ρ

(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for the system resources i in the set of local resources and wherein any given local resource in the set of local resources is comprised of a plurality of n(i) entities;

wherein the transaction log contains a set of measured resource costs R_j(t) corresponding to a group of transactions of the set of transactions occurring at a plurality of times t and for the system resources j in the set of local resources, the step of generating a system model further comprising;

a. computing from the set of system resource utilization data, a resource cost, C(i,t′

), for each system resources i in the set of systems resources according to the formula,

C(i,t′

)=ρ

(i,t′

)·

Δ

t′

·

n(i)

b. computing a total system resource cost, C_sys(i, T), for each system resource i in the set of local resources, for a given test period T, by summing the resource cost for the test period T according to the formula,

17. The method of claim 13 wherein the set of system resource utilization data contains a set of measured resource utilizations, ρ

_s(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for system resources i of the set of remote resources and wherein any given remote resource is comprised of a plurality of n_s(i) entities;

wherein the transaction log contains a set of measured resource costs, R_j(t), corresponding to a group of transactions of the set of transactions occurring at a plurality of times t and for a group of system resources j contained in the set of remote resources, the step of generating a system model further comprising;

a. computing the resource cost C_s(i, t′

) for each server s in the set of networked servers, for all system resources i associated with the set of networked servers, using the system resource utilization data and the formula,

C_s(i,t′

)=ρ

_s(i,t′

)·

Δ

t′

·

n_s(i);

b. computing total server resource cost C_s(i, T) for all system resource i in the set of system resources on each server s in the set of networked servers for a given test period T by summing the resource costs occurring within the test period T according to the formula,

18. The method of claim 13 wherein the set of system resource utilization data contains a set of measured resource utilizations, ρ

_s(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for system resources i of the set of remote resources and wherein a given remote resource is comprised of a plurality of n_s(i) entities;

wherein the transaction log contains a set of measured resource costs R_j(t) corresponding to a group of transactions occurring at a plurality of times t from the set of transactions; and

for a group of system resources j contained in the set of remote resources, the step of generating a system model further comprising;

a. computing the resource cost Cs (i, t′

) for each server s in the set of networked servers, for all system resources i associated with the set of networked servers, using the system resource utilization data and the formula,

C_s(i,t′

)=ρ

_s(i,t′

)·

Δ

t′

·

n_s(i);

b. distributing the set of transactions from the transaction log into a set of τ

(t′

) transactions occurring in the time intervals Δ

t′

,

c. determining a straight line curve fit for the resource costs C_s(i,t′

) for each server s as a function of τ

(t′

), and performing a linear regression analysis for each system resource i of the set of system resources according to the formula,

C_s(i,t′

)=m(i,s)·

τ

(t′

)+b(i,s);

where the m(i, s) is the computed slope from the linear regression analysis and b(i, s) is the computed intercept from the linear regression analysis and where a set of correlation coefficients r²(i, s) are generated for each linear regression,

d. comparing the set of correlation coefficients r²(i,s) to a predetermined threshold;

e. if a given correlation coefficient r²(i,s) of the set of correlation coefficients for a given resource i and server s is greater than or equal to the predetermined threshold, then using the straight line curve fit to estimate a resource cost for the given resource i and server s;

f. identifying a background resource cost per time interval Δ

t′

as C₀(i,s)=b(i,s); and

,

g. identifying a resource cost per transaction as R(i,s)=m(i,s).

19. The method of claim 18 including the additional step of simulating the resource cost on each of the servers in the set of networked servers as a function of a transaction throughput, τ

_. which includes the calculation of resource costs according to the formula, C(i,s)=R(i,s)*τ

+C₀(i,s), for each system resource i of the set of system resources and each server s of the system of servers.

20. The method of claim 18 including the additional step of simulating the background resource utilization on the servers in the set of networked servers which includes the calculation of background resource utilization, ρ

(i,s), according to the formula,

ρ

₀(i,s)=C₀(i,s)/T,

for test period T, for all resources i and for all servers s.

21. The method of claim 18 including the additional step of simulating the resource consumption on each of the servers in the set of networked servers by assuming that each transaction will consume R(i, s) of resource i on server s.

22. The method of claim 13 wherein the set of system resource utilization data is comprised of a set of measured resource utilizations, ρ

(i, t′

), measured at a set of times t′

over a time interval Δ

t′

for the system resources i in the set of local resources and wherein any given local resource in the set of local resources is comprised of a plurality of n(i) entities;

wherein the transaction log contains a set of measured resource costs, R_j(t), corresponding to a group of transactions of the set of transactions occurring at a plurality of times t and for the system resources j in the set of local resources, the step of generating a system model further comprising;

a. computing from the set of system resource utilization data, a resource cost, C(i,t′

), for each system resources i in the set of systems resources according to the formula

C_sys(i,t′

)=ρ

(i,t′

)·

Δ

t′

·

n(i)

b. computing from the transaction log the total application resource cost, C_app(j, T), during the test period T using the formula,

23. The method of claim 22 including the further step of submitting the average server resource cost per transaction, R_serv(i), to a discrete event simulator.