lssndrbarbieri@gmail.com
Alessandro Barbieri
LLNL/muster
The Muster library provides implementations of serial and parallel K-Medoids
clustering algorithms. It is intended as a general framework for parallel
cluster analysis, particularly for performance data analysis on systems with
very large numbers of processes.
The parallel implementations in the Muster are designed to perform well even
in environments where the data to be clustered is entirely distributed. For
example, many performance tools need to analyze one data element from each
process in a system. To analyze this data efficiently, clustering algorithms
that move as little data as possible are required. In Muster, we exploit
sampled clustering algorithms to realize this efficiency.
The parallel algorithms in Muster are implemented using the Message Passing
Interface (MPI), making them suitable for use on many of the world's largest
supercomputers. They should, however, also run efficiently on your laptop.