Method to identify performance and capacity bottlenecks of complex systems

• US 20100125755A1
• Filed: 11/18/2008
• Published: 05/20/2010
• Est. Priority Date: 11/18/2008
• Status: Active Grant

• 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
• Granted Patent: US 8,015,276 B2
• Time in Patent Office: Days
• US Class Current: 714/37
• 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 service transaction type are located;

retrieving the WFM data;

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

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;

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

determining an average Response Time Model RTM(x);

estimating response time model z₀, z₁ and z₂ parameters from the average response time RT_h and volume counts count_h data;

calculating a capacity warning valve

$\frac{1}{z_2}$

that the complex system can handle;

calculating a capacity warning value L₁;

calculating a capacity limit value L₂; and

analyzing a time trend of volume counts count_h to determine when the capacity warning value L₁ and capacity limit value L₂ will be reached if current volume counts count_h trends continue.

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 an invocation data record for “

duration”

, calculating a difference between “

start time” and

“

end time”

as the “

duration”

, wherein “

duration”

is a service transaction type invocation response time.

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

7. The method according to claim 1 wherein determining response time RT_h aggregation levels and volume counts count_h further comprises:

calculating an average response time RT_h for successful transactions within the predetermined measurement interval; and

for 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 a predetermined measurement interval;

summing all “

failed”

invocation data records within the predetermined measurement interval; and

if the “

failed”

sum is greater than a predefined percentage of the “

success” and

“

failed”

sum, 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 a 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

$\frac{\sum <br><br><br><br>x_h<br><br>y_h}{\sum <br><br><br><br>{\left(x_h\right)}^2}.$

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 a negative slope s.

15. The method according to claim 1 wherein estimating response time model z₀, z₁ and z₂ parameters further comprises modeling the RT_h of a specific function RTM(x) for a given load x as the sum of the sojourn time in each service node required to process the function as

$z_0+\frac{z_1}{1-z_2}$

x ,

wherein z₀ is the expected amount of time to traverse non-bottlenecked nodes, z₁ is the expected amount of time to traverse a bottleneck node under no transaction load and z₂ is the average amount of time of a bottleneck node per transaction type.

16. The method according to claim 12 further comprising using the data pairs (count_h,RT_h) to estimate the z₀, z₁ and z₂ parameters.

17. The method of claim 16 wherein to estimate the z₀, z₁ and z₂ parameters further comprises:

letting the z₀, z₁ and z₂ parameters be constraint values that minimize the sum of (log(RT_h)−

log(RTM(count_h|z)))²; and

using a conjugate-gradient method to find z₀≧

0, z₁ and z₂>

0.

18. The method according to claim 17 wherein when z₀ is not equal to zero, the capacity warning value L₁ is

$\frac{z_0+z_1-{(z_0}^{}}{}$

z 1 + z 1 2 ) 1 2 z 0 

z 2

and when z₀ is equal to zero, the capacity warning value L₁ is

$\frac{1}{2<br><br><br><br>z_2}.$

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

$\frac{4}{}$



z 0 + 3 



z 1 - ( 8 



z 0 

z 1 + 9 



z 1 2 ) 1 2 4 



z 0 

z 2

and when z₀ is equal to zero, the capacity limit value L₂ is

$\frac{2}{3<br><br><br><br>z_2}.$

20. The method according to claim 17 wherein z₀+z₁ is the best expected response time that can be achieved.

21. The method according to claim 1 wherein the Response Time Model RTM(x) is

$z_0+\frac{z_1}{1-z_2}$

x .