We present the parallel particle filtering (PPF) software library, which enables hybrid shared-memory/distributedmemory parallelization of particle filtering (PF) algorithms combining the Message Passing Interface (MPI) with multithreading for multi-level parallelism. The library is implemented in Java and relies on OpenMPI's Java bindings for inter-process communication. It includes dynamic load balancing, multi-thread balancing, and several algorithmic improvements for PF, such as input-space domain decomposition. The PPF library hides the difficulties of efficient parallel programming of PF algorithms and provides application developers with a tool for parallel implementation of PF methods. We demonstrate the capabilities of the PPF library using two distributed PF algorithms in two scenarios with different numbers of particles. The PPF library runs a 38 million particle problem, corresponding to more than 1.86 TB of particle data, on 192 cores with 67% parallel efficiency.

Additional Metadata
Persistent URL dx.doi.org/10.1049/cp.2014.0529, hdl.handle.net/1765/87153
Conference IET Conference on Data Fusion and Target Tracking 2014: Algorithms and Applications
Demirel, O, Smal, I, Niessen, W.J, Meijering, E, & Sbalzarini, I.F. (2014). PPF - A parallel particle filtering library. Presented at the IET Conference on Data Fusion and Target Tracking 2014: Algorithms and Applications. doi:10.1049/cp.2014.0529