by Arne Brutschy, Alexander Scheidler, Martin Middendorf, Daniel Merkle
March 2008
Submitted to ANTS 2008 conference
Table of Contents |
Runtime functions were used to generate runtimes for each job/slices pair in the system. Each function generates a set of runtimes for a single job, defining the progression of job runtime with increasing number of slices configured for this job. The generated runtimes exhibit a characteristic defined by the underlying runtime function, i.e. the exponential runtime function creates runtimes which are exponentially decreasing with increasing number of slices. Runtimes were generated as follows. denotes the class of a runtime function. Each class consists of a base function , which was used to create the function's values on an interval of . Afterwards, was used to generate random start- and endpoints (uniform distribution). Finally, the function values generated by were scaled to the random generated start- and endpoints and the required number of data values (s values, one for each slice configuration) were extracted. Table 1 lists all base functions and restrictions. Minimal runtime used in the experiments was , maximal runtime was . The required minimal difference between start- and endpoint used by some functions was set to .
The following finite state diagram represents the states of the scout units of the ant-inspired models. See caption for further details.
Figure 2 shows the optimization of the quorum threshold parameter Tfix for the fixed quorum theshold model regarding the fitness.