Case Study: Sergel Financial Services
Sergel hired Apica to assist with the analysis and performance optimization of a business-critical we portal that handles in excess of five terabytes of data constantly accessed for a range of transactions.
Technical challenge
- Can our web services handle thousands of transactions every single minute?
- How can we best configure our web application and server environment to optimize performance during high load?
- How can we verify the performance of the systems while in production?
The IT-manager of Sergel, Egon Ringden, together with Apica initiated a performance project to optimize the performance of the production environment of the web services and to verify the maximum performance of the web site. For example, tests were performed of new, but not yet deployed, web services. This test was designed to simulate various scenarios where between 1,000 and 10,000 users are logging on to a customer site, and collect information from a number of Sergel’s web services or directly from application pages.
The test result was compiled in terms of number of active users, data from server CPU and web server as well as response times for the scenario.
Examples of test results and actions
The result of the first test was that the new web service, without any kind of enhancement measurements, initially could cope with approx. 500 concurrent sessions over a single minute. Backend-servers, however, were not under significant stress; in a typical case, a full load on the front end should put the SQL-layer at 100% usage. If that correlation was missing, it could be concluded that there was a bottleneck somewhere between the web service and the database—either in the business layer or on the net.
The next action taken was to test from various segments on the net. The results were that a test from the ”inside” showed better results than a test from the “outside”. The problem was identified to be a new firewall/load balancing solution, with a default value at the maximum number of https-sessions. A simple change of the configuration eliminated the bottleneck. The new service could then be deployed delivering across the platform at full capacity.
The load test script was transformed to an “outside” check, which in production measures the response time for the web service calls.
Summary
Complex production environments will let themselves be defined only in full scale tests! Configuration problems will typically only show up under full load (which in reality only happens occasionally) but still will affect performance at lower traffic intensity. This type of system failure is hard to find in test environments. The only sure way of finding out is to perform a load test of the production environment of the system and analyze how the individual components, of which the site consists, performs.
"The integration between performance testing and application monitoring are two important components in a a process where you want and analyze how individual components perform", said Egon Rongden, IT-manager at the Sergel Group.
About Sergel Group:
Sergel offers transaction intensive businesses with high demands on quality cost efficient services within the service sectors Credit, Accounts Receivables, Collection services, Billing and CRM. Among Sergel's customers are leading companies within telecom, energy and real estate businesses. Today, Sergel Group is a leading company in its business segment, with approx. 375 employees and offices in Sweden, Norway, Finland, Lithuania and Latvia.
For further information, please contact:
Sven Hammar
CEO, Apica AB
Phone: 46 (0)8 400 27 319
Mobile: 46 (0)735 - 05 90 68
E-mail: sven.hammar@apicasystem.com
and
Egon Rongden
IT-manager, Sergel Group
E-mail: egon.rongden@sergel.com