- Auto/manual correlation
- Test Analysis
- Test script customization
- Test scheduling
Actually we’re implementing and consulting with our customers in terms of performance testing using LoadRunner or Performance Center.
For example, we’re in the middle of a project for implementing a distributed Performance Center system, with six load generators for testing an Oracle ERP system that holds about 800,000 users, with an average annual increase of 40,000 users.
Our target is to implement Performance Center, integrate it with HP ALM, design the load tests for simulating 70K virtual users that are hitting the system with different scenarios, and finally we will present reports to ERP management staff about the response trend of the system, such as virtual users passed/failed, server throughput/sec, transactions taken time etc.
I think there is a lot of configuration and customization that can come in for optimizing the recorded load test in terms of CPU and memory used and, accordingly, the service/process that are running this v-user script at load generator. This would be very helpful especially when running a huge number of v-users simultaneously.
I've used it for two years.
Running virtual users especially with a huge number is a very challenging task. The load generators may over utilize its resources (CPU/ memory) due to improper default configurations, or improper use of LoadRunner and Performance Center. Many considerations should be taken into account when optimizing resources when doing this task. For example, running the recorded script as a service at the load generator is different from running it as a process.
Also, customizing the auto-generated script in virual user generation may cause problems when ramping up the number of users. For example, simulating a generated list of variables that would be used by every virtual user is important.
Lastly, taking care of variable sizing and limitation should be reviewed, and otherwise errors may appear at certain point of running the script.
Above moderate, but HP is trying to enhance its support level.Technical Support:
Above moderate, but HP is trying to enhance its support level.
The initial setup is straightforward when following the installation guide steps.
I did the implementation by myself as part of a project for one of our customers. My advice is to understand the business need that would guide to a proper scope and design guidelines. Performance Center can be implemented in different ways (standalone, distributed, with/without integration with HP-ALM or HP Diagnostics, number of load generators and their location etc.). The implementor should arrange for the best and quickest setup that will fulfill the goals, otherwise the implementation would end up as a trip of trial and error, and probably fail, or exceed the time plan.
Our customers think of the following benefits as ROI of using HP Performance Center:
The license of Performance Center is not cheap and may be very expensive for some protocols, such as SAP and Oracle. Accordingly, optimizing the license to what is only needed – in terms of protocol and number of virtual users can save a lot of money. When the load testing is limited with a start and end date plan, I recommend using temporary licenses, or a pay-as-you-go model of license, it may be more expensive in the short run, but of course it would be more economical in the long run.
From outside HP no, but inside HP there are two solutions for doing the load testing; LoadRunner, and Performance Center. The concept is almost the same, but Performance Center excels in big projects, and working with different teams besides scheduling tests feature.
My advice is to identify well the scenarios, protocols and maximum number of virtual users needed for load testing. Also, simulating a real-world load testing scenario is very important if you need to get a near-real results. For example, simulating network speed to reflect the real case. Some scenarios may require using a paid e-service, which may cost a lot when simulations repeat with a huge number of virtual users. At this point, another product called HP Service Virtualization could be used to sniff the in/out going traffic, then simulating the e-service function later after learning its function. At this point we can then replace the real e-services communications with HP Service Virtualization, so a budget-wise trade-off may be held between using HP Service Virtualization and the paid e-services. I recommend consulting HP presales to get the most proper setup with least possible licensing.