Hotsos Symposium Speaker – Neil Gunther

Neil Gunther is a returning Hotsos Symposium presenter.


Neil GuntherNeil J. Gunther, M.Sc., Ph.D., is an internationally known performance researcher and consultant who founded Performance Dynamics Company in 1994. He is the author of 3 books and many technical papers and presentations on performance analysis and capacity planning. Dr. Gunther received the prestigious A.A. Michelson Award from CMG in 2008, and was elected a senior member of both ACM and IEEE in 2009.

Presentation Title

Time Bandits: How to Analyze Fractal Query Times


No doubt you are familiar with fractals as those infinitely recursive, self-similar, geometrical patterns made famous by the late Benoit Mandelbrot. Fractals are power laws in SPACE. It turns out that fractals can also impact computer system performance, but to understand how that works we need to consider fractals in TIME. That's because time is the zeroth-order performance metric.

Time-fractals lead to self-similar clustering of event-instants that can significantly impact the performance of things like the packet latency in data networks and (so it seems) queries in databases. This fractal clustering in performance data is a signal that undesirable large-scale correlations are at play.

Since fractals are power-law distributions with potentially infinite correlations, they do not conform to standard statistical distributions (e.g., normal, Poisson, exponential). So, how can we possibly analyze such fractal performance data?

In this talk, Dr. Gunther will show you how to detect and analyze fractal performance data. In particular, he will:

  • Explain simply the origin of fractals and the term "power law"
  • Present two case studies where fractal effects were at work in Oracle data
  • Discuss the pitfalls of misinterpreting power law effects

No previous knowledge of fractals or power laws will be assumed in this presentation.

Presentation Materials

Presentation materials are available online to attendees only.


The speaker schedule is as follows: