Experiment configuration. We utilised the example presented in Section 22.6.1 for evaluation. Bidders’ reserved capacities for price and execution time for individual component services were picked randomly from the fixed intervals as presented in Table 4.2. And bidders were willing to give discounts when they bided for combinations of component services. In the experiments, some bidders would bid for the combination of services B and C, and some for the combination of services D and E (see Figure 2.2). Considering that the varying degrees of complementarity of the component services lead to different maximum discounts, we modelled the maximum discounts that the bidders were willing to apply to combinations of component services with small, medium and large degrees of complementarity using beta distribution functions presented in Figure 4.5. The Chi-Square distribution model presented in Figure 4.4 was used to compute the dynamic minimum increments and the ask QoS. Table 4.2 Intervals for bidders’ reserved capacities Figure 4.5 Maximum discounts for component services with small, medium and large degrees of complementarity We conducted four sets of experiments with zero, small, medium and large degrees of complementarity assigned to the combinations of services B and C, and services D and E, respectively. With zero complementarity assigned to the combinations of component services the combinatorial auction is actually equivalent to traditional single-item auctions because no discounts or premium would be applied by the bidders. Random values from the intervals presented in Table 4.2 were picked to specify the reserved capacities of the bidders. The QoS proposed in a bidder’s first bid (the bid proposed in the first round as response to the request for proposal) is selected from a reasonable range conforming to the normal distribution. For example, if a bidder’s reserved capacity for the price of the rough processing service is $100,000, then the price proposed in its first bid will be selected between [$150,000, $200,000]. This configuration captures the nature of the bidders with various concession spaces. To access the performance of CASS in environments on different scales, we conducted each set of the experiments with 25, 100 and 250 bidders. By doing so, an average number of 5, 20 and 50 bidders competed for each service, creating different competitive environments. In each experiment, a random proportion (between 30% to 50%) of the bidders was given the ability to bid for com...
Experiment configuration. ID FR004 Requirement Uniform experiment description Comment An experiment requires a standardised representation within the information model of the IoT Lab platform Attributes Need v.high Stability v.high Priority v.high
