Customer Reviews

Guerrilla Capacity Planning: A Tactical Approach to Planning for Highly Scalable Applications and Services

5 star:		(4)
4 star:		(2)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

Neil Gunther has undertaken an important work, that of teaching to IT professionals the basics of measuring and modeling the scalability of parallel computer systems. The model that he develops in his book is a useful starting point; however, this model fails to provide a sufficiently general basis for modeling the behavior of the wide variety of extant parallel computer systems.

The "universal scalability model" that he describes in section 4.4, and for which he provides figure 4.8 and equation 4.31, extends Amdahl's Law via the addition of a "coherency" term that models effects such as data exchange between parallel processes. And although Gunther suggests that this coherency term ought to grow linearly with the number of parallel processes, and hence should appear as a quadratic term in equation 4.31, this coherency term depends on the specific communication architecture of the computer system and can grow non-linearly.

For example, in a distributed memory architecture such as a cluster of compute nodes, coherency effects may grow linearly with the number of processors due to communication between the compute nodes. However, in a shared memory architecture, coherency effects may grow by the logarithm of the number of processors because one processor may not communicate directly with all other processors. Instead, one processor may send information to two other processors, and each of those two processors may send information to two more processors, and so forth. Therefore, in order to model the communication that occurs in a shared memory architecture, the quadratic n(n-1) coherency term in equation 4.31 should be replaced by a n*log(n) term.

In addition to the above example, I have obtained performance data for other parallel computer systems for which the universal scalability model fails to model adequately the scalability for a variety of different reasons. Thus, although Gunther's book is a useful introduction to the subject of measuring and modeling the behavior of parallel computer architectures, his universal scalability model should not be considered to be universal. Instead, that model is a useful starting point for modeling the behavior of a parallel computer architecture.

I've read the other reviews and they seem to ignore the "Guerrilla" concept. The fact that scientific analysis is ignored and decisions made on perceived knowledge in most companies for me is the key to the book. Excel is a great way to get the performance point across even with precision errors. Getting management buy in is 99% of the process. GCP makes that argument simple. Read this book and get the word out. Performance is not linear!

Very readable coverage of Capacity Planning and Performance Management. Doesn't presume any previous knowledge, but doesn't talk down either. Several good chapters talking about queueing theory.
A great practical handbook.

First of all, this book was worth the money I spent on it. I came away from reading this book with a clear understanding of the differences between speed and scale, and with a system for modelling the scalability of systems in general.

However... really all of this value was in the first quarter of the book. I read on and read on looking for further conceptual gems but they weren't to be found.

I guess that books are "meant" to be at least a particular length, but this one could have been much shorter and more concise.

Alright I finally got the book! As promised a proper review.

This is the third Gunther text I have purchased. I believe his work is unique and is filled with interesting ideas. Some of the ideas he takes from his previous works but overall it is a nice compliment to his current bibliography.

If you are interested in performance tuning and capacity planning this is a good purchase. I highly recommend it. Plus I love the little "Guerilla Manual" that is included in the back sleeve of the work.

With Wall Street analysts drives the planning horizon, Management prefers getting a sense of direction quickly and repeatedly, instead of belated precise readings of compass bearing. It is in this agile and opportunistic spirit and philosophy that Dr. Gunther introduces Excel, linear regression, and 2 parameter scalability models into the performance analysts' tool chest.

Excel is ubiquitous. It is also easy to use. Use it. If there is sufficient time, better tools such as R or Mathematica can be used to cross-check Excel results. Similarly, linear regression is another tool in the agile performance analysts' tool chest.

Two chapters I have not seen presented elsewhere are the virtualization spectrum and effective demand. In a prior job, having virtualization spectrum chapter available to me would have save me much grief with an workload manager. The effective demand makes another useful capacity project tool to keep handy.

The best part is Dr. Gunther's 2 parameter universal scalability model. It can be immediately used to frame your load testing results to project application scalability. This alone is worth the cost of the book and admission to his classes.

Conjecture 4.1 on page 65 on 2 parameters are necessary and sufficient for scalability model based on rational functions are an interesting open questions. Given that the denominator is a quadratic equation with c = 1, we should be able to argue that it behaves like a parabola, except with c = 1, we won't get into singularity/infinity. For more details, please see Dr. Gunther's blog at

[...]