Newsletters

Using a several step process, Hewlett-Packard Company’s Partner Technology Access Center (PTAC) used TeamQuest software to successfully collect data, model, do what if scenarios, and create a sizing guide. The application sizing information provided this TeamQuest software user to recommend the best hardware to her customer.

Application Sizing Service Case Study
HP consultant Karen Karwoski, along with PTAC members, worked with an ISV that requested a sizing guide for its multi-tiered customer billing application. It had a web server tier, an application tier and a database tier.

The ISV agreed to set up a typical application environment for testing. Application tuning was performed prior to running tests for collecting data for modeling. Business metrics were defined so that the application workload could be correlated to system resource utilization – in this case, thousands of customers processed per hour.

Data Collection for TeamQuest Model included instrumenting OVPA files to ensure that the data was complete for modeling. The ISV agreed that the duration of the tests would be at least one hour to make sure HP had a good representation of the workload. This test, it turned out, ran to one hour and twenty minutes.

The next issue was whether to model each tier individually or to use the multi-tier feature of TeamQuest Model to address the environment as a whole. In the end, PTAC chose a hybrid approach – modeling each tier individually and then once completed, use multi-tier modeling to validate the findings.

This meant building a model for each tier, calibrating the model, consolidating workloads on each tier so that one workload included all customer billing processes, associating the business units with the system resource utilization of the customer billing workload at each of the tiers, validating each of the models, and making projections for defined systems at each tier.

The ISV provided important criteria such as a maximum CPU utilization of 80 percent for each tier. The customer also defined its preferences for each tier in terms of types and sizing of systems.

Revelations
TeamQuest Model revealed the following:

Web tier: the web server was a 44 CPU HP Superdome. By ramping up the amount of customers processed per hour in a series of 10 increments, we discovered that the 80 percent utilization threshold occurred at Step 9 at a throughput of 1125K/hour.

Application tier: the application server was a 12 CPU Superdome. 80 percent CPU utilization took place between Step 3 and Step 4. As the throughput for Step 3 was 563K and for Step 4 was 656K, the throughput at 80% was about 600K/second.

Database Tier: The database server was also a 12 CPU Superdome. This time, the utilization of the actual test was done at about 80 percent. As a result, we projected downward to see what the behavior of the application would be. Throughput on the database server at 80 percent was approximately 375K/hour.

To validate these three separate tiered models, the PTAC created a multi-tiered model in TeamQuest Model using the same raw data. Once created, several different scenarios were set up for validation purposes. For example, if the ISV wanted to create a multi-tiered environment for a maximum of 600K customers/hour, that meant using a 24 CPU Superdome at the web server layer, a 12 CPU Superdome at the application layer and a 24 CPU Superdome at the database layer. This conformed well to the individual tier models conducted using TeamQuest Model.

TeamQuest Model also indicated that a disk I/O bottleneck was occurring. By adding more controllers and disks at certain data levels, this problem was averted.

In order to have the data necessary to create an application sizing guide, what-if scenarios were run on various system types and sizes at different throughput levels. By running these modeling scenarios, we could determine what the maximum throughput levels (thousands of customers processed per hour) were on all of the defined system types and sizes. This then allowed us to create an application sizing guide that could be used by the sales team to provide initial system sizing information to customers.

The ISV then received a report stating an overview of the testing environment, a summary of tuning performed during the test, an identification of potential system bottlenecks, a description of the tests performed and the results, and an overview of the modeling process and projections. Most important, it provided detailed application sizing results in table format to give the ISV the ability to confidently recommend HP hardware to its customers.