Review of Configuration and Capacity Planning for Solaris Servers

Configuration and Capacity Planning for Solaris Servers
Brian L. Wong
Prentice Hall

Reviewed by Nick Christenson,

February 24, 1998

There are several good books available for system administrators who want to know how to solve specific problems that they may encounter, but there had been a notable lack of information on how to specify and build systems to begin with. As the title suggests, Configuration and Capacity Planning for Solaris Servers attempts to do just this, providing information on how to determine what gear to obtain and how it should be set it up to achieve certain ends.

The book is divided into nine chapters, the first of which is titled Methodology and defines terms and concepts that will be used later in this book. It's a pretty decent introduction, as it is absolutely necessary that the reader and author agree on the terminology to be used.

The next four chapters discuss configuring different kinds of servers using Sun gear, although this information is immediately applicable to at least other Unix system. Advice on designing and benchmarking NFS file servers, database servers, various flavors of Internet servers, and timeshare systems are all discussed here. Wong advises the reader on how to estimate various kinds of load, and then what (Sun) gear to buy in order to achieve these ends. This is remarkable. As far as I know, this is the first time such a comprehensive guide to determining resource needs has appeared in print. Wong's advice on how to benchmark these systems and how to design a system to meet these needs are generally good, or at least I agree with most of them. For example, the author and I share opinions on the importance of small disks as opposed to large disks for file system throughput, the common overemphasis on CPU speed, and other system design principles that have rarely been published.

On the other hand, once the author begins discussing specific cases and recommendations, I start to have disagreements with his conclusions. Just one example is on the configuration of Usenet News servers, a topic I happen to know a bit about. One simply can't make good recommendations about what sort of gear one needs without knowing how long expire times are going to be, at which speeds clients connect, what a typical client's reading profile looks like, etc.. Making recommendations without this information is, in my opinion, simply foolish, and Wong does this in several areas, often using logic which I cannot agree with. Despite this, if one is incapable, for whatever reason, of calculating these metrics themselves, then I'm willing to concede that Wong's numbers would be considerably better than nothing, but I urge folks to take these numbers with a healthy dose of salt. In any case, I'm fairly sure that the author would agree that one is always much better off doing all the requirements gathering themselves rather than depending on anyone's numbers.

This leads,though, to another problem I have with the book. While I expect a book written by a Sun employee published by Sun Microsystems Press to be pro-Sun, in my opinion this occasionally gets out of control. For example, how many folks need advice on how to use a Sun box as a high speed router? Sure, sometimes they come in handy as low speed gateways, and I admit that I know a couple of places where they do mess with gated, but if your Sun runs out of capacity as a gateway, the best advice for almost site is, "buy a bigger dedicated router." Similarly, even at the time the book was written, if "Buy a SPARCserver 1000" is the correct answer, I can't imagine what the question could possibly be. It seems that the author was reaching for a reason for someone to buy each piece of gear in Sun's catalog at the time. In one case, the application requires low performance but several SBus slots. Sure, one could do this with a SPARCserver 1000, but it would be cheaper to do it with multiple Sparc 5s or 20s.

Another related complaint is that at the time the book was written, Sun was overhauling its product line with the release of the UltraSparc chips. A lot of the recommendations in this book were wildly out of date at the time of its publication. Also, no thought was given to weighting recommendations by architecture lifetime, and this, in my opinion, undermines the author's credibility in making other recommendations. For example, even if I didn't need the horsepower now, I'd have to recommend the purchase of an Ultra 1 over a Sparc 10, even at slightly higher cost, because the product has a longer life. Because of this, perhaps surprisingly, this book may be more valuable to folks to work in non-Sun shops who would not be tempted to take this advice on its face.

The last four chapters are on Sun system architecture, storage system architecture, backup systems, and Solaris 2. These chapters are all extremely good. This book contains the best single, simple source of information explaining the various available RAID systems including their relative strengths and weaknesses (although I have some minor quibbles with a little of the information) and tape backup systems I've seen. In fact, these chapters make the book worthwhile by itself.

It sounds like I'm being very hard on this book, and I am, but it's not because I think that the book shouldn't be read. On the contrary, I think it's one of the more important books for system administrators written in the last few years. It is because this book is so important that I'm being so critical. This is the first book of its type, and it is well worth reading, I just feel that it should be read critically. It should also be pointed out that the topics of the book intersect sharply with my areas of expertise and areas of strongest opinion, so others may not have as harsh a reaction as I had.


Despite some weak points, a stronger bias toward weaker Sun solutions than is warranted and some questionable advice on some specific implementation recommendations, this is a very important book that covers much ground for the first time anywhere. The sections explaining how systems work and on how to measure requirements are extremely good, but the specific recommendations are both out of date and should be viewed skeptically in any case. Nonetheless, I recommend the critical reading of this book by everyone involved in the design and specification of computing systems.

Note: I received a copy of this book from Sun free of charge. Other than this, I have no interest, financial or otherwise, in the success of this book.