Massive-Scale Gaze Analytics Exploiting High Performance Computing

Methods for parallelized eye movement analysis on a cluster are detailed. The distributed approach advocates the single-core job programming strategy, assigning processing of eye movement data across as many cluster cores as are available. A foreman-worker distribution algorithm takes care of job assignment via the Message Passing Interface (MPI) available on most high-performance computing clusters. Two versions of the MPI algorithm are presented, the first a straightforward implementation that assumes faultless operation, the second a more fault-tolerant revision that gives nodes an opportunity of communicating failure. Job scheduling is also briefly explained.