Preloading policies for the Virtual Computing Lab

In the Virtual Computing Laboratory (VCL) users request a computing resource (e.g., a blade) and customized software (an “image”). The software is loaded onto the blade from an image server. If loading is performed at the time of the request, the user experiences a delay before s/he can access the blade. This delay may be several minutes long, depending on the type of image requested. A “preloading policy” aims at reducing this delay; the basic idea is to anticipate what images may be requested in the future and start loading them onto free blades in advance of reservations. In this paper, we define several preloading policies and show via simulation that they can significantly reduce user-experienced delays. We drive the simulation with two sets of workloads: (a) actual traces derived from historical data collected over several years of operations at North Carolina State University’s VCL (approximately 155,000 reservations), and, (b) synthetic workloads. Performance metrics include user-related ones (e.g., waiting time histograms and averages) and system-related ones (e.g., blade/image utilization). We use the simulator to address “what if”, system design questions and select system configuration parameters.