Comments on: Computer processes, human processes, and scalability http://www.johndcook.com/blog/2008/07/15/computer-processes-human-processes-and-scalability/ The blog of John D. Cook Sat, 11 Feb 2012 22:42:11 -0500 http://wordpress.org/?v=2.8.4 hourly 1 By: Neil Gunther http://www.johndcook.com/blog/2008/07/15/computer-processes-human-processes-and-scalability/comment-page-1/#comment-16962 Neil Gunther Tue, 05 May 2009 05:57:16 +0000 http://www.johndcook.com/blog/2008/07/15/computer-processes-human-processes-and-scalability/#comment-16962 A lot of HPC codes can exhibit near-linear scalability if the computational tasks are sufficiently fine-grained and can be multi-threaded. But <a href="http://perfdynamics.blogspot.com/2009/02/poor-scalability-on-multicore.html" rel="nofollow">not always</a>. Note that the Sandia representation of latency (the bathtub curve) is analogous to <a href="http://en.wikipedia.org/wiki/Brooks'_law" rel="nofollow">Brooks' law</a> for the "scalability of human processes." :) As a mathematician, you might find it interesting that scalability can be <b>quantified</b> as <a href="http://arxiv.org/abs/0808.1431" rel="nofollow">"A General Theory of Computational Scalability Based on Rational Functions"</a>. This unifying model subsumes Amdahl's law, Gustafson's law and the retrograde effects seen at Sandia (as well as elsewhere). A lot of HPC codes can exhibit near-linear scalability if the computational tasks are sufficiently fine-grained and can be multi-threaded. But not always. Note that the Sandia representation of latency (the bathtub curve) is analogous to Brooks’ law for the “scalability of human processes.” :)

As a mathematician, you might find it interesting that scalability can be quantified as “A General Theory of Computational Scalability Based on Rational Functions”. This unifying model subsumes Amdahl’s law, Gustafson’s law and the retrograde effects seen at Sandia (as well as elsewhere).

]]>