2017 OLCF User Call: R and pbdR

The pbdR project enables native and low overhead use of HPC scalable libraries in R language data analysis workflows. Distributed parallel dense linear algebra, which is essential in complex data analysis, is available without changes to serial R code. Data distribution and redistribution functions enable in-memory distributed data wrangling for workflow optimization. Building of custom analytics is further facilitated by high-level use of MPI collectives, where most rank wrangling is automated without user involvement, and by the availability of profiling facilities that range from PAPI library hardware counter information to instrumented MPI communications. Recent additions include a client-server interface for interactive SPMD style parallel code development. We will overview pbdR on this call.