Abstract
In this report, I describe the design and implementation of an inexpensive,
eight node, 32 core, cluster of raspberry pi single board computers, as well as
the performance of this cluster on two computational tasks, one that requires
significant data transfer relative to computational time requirements, and one
that does not. We have two use-cases for the cluster: (a) as an educational
tool for classroom usage, such as covering parallel algorithms in an algorithms
course; and (b) as a test system for use during the development of parallel
metaheuristics, essentially serving as a personal desktop parallel computing
cluster. Our preliminary results show that the slow 100 Mbps networking of the
raspberry pi significantly limits such clusters to parallel computational tasks
that are either long running relative to data communications requirements, or
that which requires very little internode communications. Additionally,
although the raspberry pi 3 has a quad-core processor, parallel speedup
degrades during attempts to utilize all four cores of all cluster nodes for a
parallel computation, likely due to resource contention with operating system
level processes. However, distributing a task across three cores of each
cluster node does enable linear (or near linear) speedup.
Users
Please
log in to take part in the discussion (add own reviews or comments).