262 CASE 12 User-Centered Design for Middleware The First Usability Study While the company was growing, the user experience team was also expand- ing. Carl hired two summer interns whose help allowed him to do more user testing in the summer of 2005 than would have been possible otherwise. Besides the inventory, there were several other new features that needed to be tested. The two major and very high-value concepts being introduced in VirtualCenter 2.0 were failover and automatic load balancing. Failover is a concept that allows the system administrator to specify what to do in the event that a server goes down while running virtual machines. This concept was not new to these users. Dealing with server failure is what they do all day, everyday, whether they use virtual machines or physical ones. The expectations for this feature were high, and the interface had to meet all the users' needs. Automatic load balancing allows the system administrator to group a set of servers into a cluster so they can share resources and, using VMotion, automatically move running virtual machines from an overloaded host to a more idle one. Users do not have this ability at all with physical machines, and with VirtualCenter 1.0 could only migrate virtual machines manually with limited insight into the overall performance of the hosts so they could make load-balancing decisions. Before introducing this new concept, Carl wanted to learn whether users would understand the value of this feature and whether it would integrate well into the existing product. The Protocol Carl called Pamela again to help conduct a usability study. Pamela put together a protocol, and it quickly became apparent that it was going to take the user several hours to complete the necessary tasks to get the feedback the team needed. Pamela knew from past experience that users can be expected to be at their best only for about 90 minutes, at most 2 hours before they become fatigued and less effective and focused. Based on her advice the team decided, with much reluctance, that the study would need to be limited in scope and some of the features would need to be left untested. It was a very difficult decision to choose which features to leave out of the test, but ulti- mately the test included what were perceived to be the two riskiest features--the new inventory model and load balancing. The Prototype Carl and the user experience team developed a medium-fidelity prototype by taking screenshots of VirtualCenter 1.2 and modifying them to demon-