Method to identify performance and capacity bottlenecks of complex systems

• US 8,015,276 B2
• Filed: 11/18/2008
• Issued: 09/06/2011
• Est. Priority Date: 11/18/2008
• Status: Expired due to Fees

• Current Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Original Assignee: AT&T Intellectual Property I LP (AT&T, Inc.)
• Inventors: Hoeflin, David A., McCaskey, Harold Orr III
• Primary Examiner(s): Nawaz; Asad M
• Assistant Examiner(s): Ashraf; Waseem
• Application Number:
  US12/313,236
• Publication Number:
  US 20100125755A1
• Time in Patent Office: 1,022 Days
• Field of Search: 709223-224, 709/226, 703/22
• US Class Current: 709/223
• CPC Class Codes: G06F 11/3409 for performance assessment
• CPC Class Codes: G06F 11/3419 by assessing time
• CPC Class Codes: G06F 11/3433 for load management allocation of a server based on load conditions G06F9/505; load rebalancing G06F9/5083; redistributing the load in a network by a load balancer H04L67/1029
• CPC Class Codes: G06F 11/3452 Performance evaluation by statistical analysis
• CPC Class Codes: G06F 2201/87 Monitoring of transactions
• CPC Class Codes: G06F 2201/88 Monitoring involving counting

Abstract

Systems and methods are described that analyze complex systems and identify potential performance bottlenecks that may affect capacity and response time. The bottlenecks are identified to resolve problems originating at a specific subsystem(s).

Claims

1. A method for determining capacity in complex systems by setting capacity values for service transaction types that assist in identifying performance bottlenecks comprising:

listing the service transaction types;

identifying Workflow Manager (WFM) data where invocation data records for each of the service transaction types are located;

retrieving the WFM data;

parsing the WFM data to acquire the invocation data records for each of the service transaction types;

for a given transaction type;

determining average response time RT_h aggregation levels and volume counts count_h, wherein the volume counts count_h is the traffic load and is a count of successful transactions within a predetermined measurement interval h and the average response time RT_h is calculated for successful transactions within the predetermined measurement interval h;

performing a linear regression of data pairs (count_h, RT_h);

determining a Response Time Model RTM (x) to be

2. The method according to claim 1 wherein the service transaction types comprise “

get home address”

, “

get billing address” and

“

get billing information”

.

3. The method according to claim 1 wherein retrieving WFM data further comprises using agents, bots, spiders and other network mechanisms.

4. The method according to claim 1 wherein parsing the WFM data further comprises obtaining invocation data records “

start time”

, “

end time”

, “

success” and

“

fail”

.

5. The method according to claim 4 wherein if the parsed WFM data does not include the invocation data record for “

duration”

, calculating a difference between “

start time” and

“

end time”

as the “

duration”

, wherein “

duration”

is the service transaction type invocation response time.

6. The method according to claim 5 further comprising filtering the invocation data records from the parsed WFM data.

7. The method according to claim 1 wherein determining the average response time RT_h aggregation levels and the volume counts count_h further comprises for the predetermined measurement intervals which have less than a predefined number of successful transactions, excluding those records from consideration.

8. The method according to claim 7 wherein the predefined number of successful transactions is at least 30.

9. The method according to claim 1 further comprising identifying abnormal complex system behavior.

10. The method according to claim 9 wherein identifying abnormal complex system behavior comprises:

summing all “

success” and

“

failed”

invocation data records within the predetermined measurement interval;

summing all of the “

failed”

invocation data records within the predetermined measurement interval; and

if the sum of the “

failed”

invocation data records is greater than a predefined percentage of the sum of the “

success” and

“

failed”

invocation data records, declaring an abnormal condition.

11. The method according to claim 1 wherein the linear regression further comprises:

calculating a slope s of a least squares regression; and

determining whether the slope s is negative.

12. The method according to claim 11 wherein calculating the slope s of the least squares regression further comprises:

for the data pairs (count_h, RT_h);

averaging all of the count_h observations as x;

averaging all of the RT_h observations as y;

calculating centered counts variables x_h equal to count_h−

x;

calculating centered average response time variables y_h equal to RT_h−

y; and

calculating the least squares regression slope s as

13. The method according to claim 12 further comprising if the least squares regression slope s is negative, stopping further analysis.

14. The method according to claim 13 further comprising examining the invocation data records producing the negative slope s.

15. The method of claim 1 wherein estimating the response time model z parameters z₀, z₁ and z₂ further comprises:

estimating the z₀, z₁ and z₂ parameters as the constrained values of z that minimize the sum of (log(RT_h)−

log(RTM(count_h|z)))² subject to the constraints z₀≧

0, z₁ and z₂>

0 using a conjugate-gradient method.

16. The method according to claim 15 wherein when the z parameter z₀ is not equal to zero, the capacity warning value L₁ is

17. The method according to claim 15 wherein when the z parameter z₀ is not equal to zero, the capacity limit value L₂ is

18. The method according to claim 15 wherein z₀+z₁ is the best expected response time that can be achieved.