June, 2010
by Mo Marikar
Performance testing of peak hourly volumes becomes the input to the sizing of servers for capacity analysis. A simple performance model the automation specialist could use to determine the theoretical peak hourly distribution would help in this effort. If this model can be used by management, then quicker acceptance would be the low hanging fruit.
Introduction
Servers in financial services production environments have peak hour volumes increase due to changes in business inputs, environment and other variables. The Performance/Capacity Analysis group has to come up with increased volumes testing prior to new releases. How one models the performance test can simplify analytics, such as how will the increase in volume of one component impact the whole.
Performance automation packages, such as HP's LoadRunner, utilize parameter files to implement variables involved in the automation. This paper illustrates the use of an excel model that could be used for changing the percentage distribution and total volumes in existing regression test suites.
Performance tests are iterated over a one hour period using unique multiple virtual users and iterations. Consecutive iterations by the same user are delayed by pacing subsequent iterations with a suitable delay. Thus in 1 hour by having pacing set to 900 seconds for example there will be four end to end transactions initiated for a single user.
Consider the case for determining how to send 2800 files in an hour using different protocols, such as HTTPS, FTPS, SFTP, VAN, SWIFTNET and AS2, in the busiest hour keeping the percentage distributions at 24, 17,20,10,11 and 18. The number of unique users available for the performance environment is 63, 51, 51, 1, 10 and 5.
Then consider how to use the model to increase this volume to 4000 files in an hour and alter the % distribution to 28,22,23,7,7 and 13.
The gist of this paper is to demonstrate the KISS principle in automation. The modeling tool used is Microsoft Excel while the implementation used was HP's LoadRunner, though it could as easily be implemented using any similar tool.
The Model
Neil Gunther has demonstrated the PDQ (Pretty Damn Quick) analyzer as a methodology in The Practical Performance Analyst [1]. The Financial Services Performance Analyst of today has to present quick models to be implemented by automation specialists. This paper illustrates the use of a spread sheet such as Excel that can be used to build that model.
Spread-Sheet Model:
Column A (Type) gives the protocols used in sending the files. Column B (User) is the number of users per protocol and is fixed. Column C (Pacing) is the time between iterations per user and is a variable that can be manipulated by the analyst. The values in Column D (The number of files to be sent in an hour per user) are equal to an hour in seconds divided by the corresponding value in Column C. Values in Column E (Total number of files in an hour) equal the values in Column D multiplied by the corresponding value in Column B.
Columns F through K give the percent of each protocol used by the total files sent by dividing the value in Column E by the grand total of files sent in an hour (Row 9, Column E) and expressed as a percent.
Thus the new model was formed by manipulating the pacing parameter as shown
Conclusion
Simplicity in modeling is key to an effective analytical model for simulating changes to peak volume distribution since management can utilize it and, more importantly, understand it.
References: The Practical Performance Analyst. By Neil J Gunther. ISBN 0-595-12674-X
[1] The Practical Performance Analyst. By Neil J Gunther, ISBN 0-595-12674-X