TCP/IP Network Administration

From Barnes & Noble

Fatbrain Review

If you are considering setting up and running a TCP/IP network for the UNIX system, there is no other book like TCP/IP Network Administration. The second edition has been updated to include information on advanced routing protocols such as RIPv2, OSPF, and BGP. The book starts off with the fundamentals of TCP/IP, overviews the history, and describes the structure of the protocol architecture, how they work, and the relationship between clients and server systems. It covers Ethernet, SLIP, PPP, Bootstrap protocol (BOOTP), dynamic host configuration protocol (DHCP) servers, NFS, NIS, and IMAP. There are chapters on troubleshooting TCP/IP and network security. The book covers Linux, BSD, and System V TCP/IP implementations. If you got the first edition, keep on the edge of networking with the latest one.

Booknews

The TCP/IP protocols (also called the "Internet protocols") are the glue that connects most UNIX networks. The approach here is practical: how to put systems on the net and keep them running--and that's what many system administrators must do these days to keep up with the times. Annotation c. Book News, Inc., Portland, OR (booknews.com)

Don Bryson

Don't Lose the Tree in the Forest

In 1991, when Hunt wrote the first edition of TCP/IP Network Administration, TCP/IP nodes were scattered across the country like occasional trees dotting a rolling cow pasture. While about 600,000 computers were on the Internet in 1991, most UNIX systems were host-centric -- one central computer serving only terminals and printers via RS232. Also, among the connected computers, TCP/IP only gave us boring things like e-mail, ftp, and telnet. There wasn't a virtual world of animated graphics and hi-fi sound to lure the game-drunk masses to the Net. Networked home PCs were rare enough to warrant raised eyebrows. The relatively few networked PCs were not using TCP/IP. Netware was king in the DOS world and UNIX-to-UNIX-Copy (UUCP) ruled the underworld of UNIX. Like the few TCP/IP nodes themselves, Hunt's subject was one lone tree in a pasture. You actually knew what to expect when you bought the first edition -- a book describing TCP/IP on a UNIX system.

The world changed. A book on TCP/IP can now discuss CGI, HTTP, NT, HTML, WINS, and a multitude of other letters that have absolutely nothing to do with administering a UNIX network. Now, installing TCP/IP can range from a thirteen-year-old installing Netscape on her father's computer to a college network administrator connecting heterogeneous UNIX machines to heterogeneous networks. A rain forest of trees sprang up around Hunt's one lone tree.

Hunt might have lost sight of his tree in such a forest. Thankfully for the rest of us, he resisted. The second edition is only an update of the first. He discusses sendmail 8.8.5 instead of version 5.6.5. He also discusses the latest versions of bind, Linux, and Solaris. He added Open Shortest Path First (OSPF) and Border Gateway Protocol (BGP) to the routing section. The information is updated, but his style and craft are the same. His focus is still sharp, his writing is still clear, and his organization is still logical. While the book has no practical value to the UNIX-free, it quietly replaces its predecessor in the UNIX required reading list for the rest of us.

Hunt is very practical in his advice. This book is a handbook for the handyman, not the speculations of a scientist. For example, when discussing troubleshooting, Hunt gives the following advice: check the cables, keep good records, keep an open mind, duplicate the problem, and keep your speculations to yourself. It sounds like he has done this before. Hunt's greatest strength is his practicality borne from years of experience.

Hunt starts briefly with the theoretical, but spends the majority of the book on the practical execution. This structure works very well. You get an overview of the subject before learning a selected tricks. It's always good to go once around the elephant before sticking your hand under its foot to remove a thorn. Knowing the size of an elephant is important if your hand is going under its foot.

The first three chapters introduce key concepts like IP headers, IP addresses, subnetting, ports, and services. The next chapter describes planning a network. Hunt discusses the practical issues involved in planning a network: Use hostnames that are easy to spell, use a mail server to collect mail, use an installation planning sheet to document the network, and reorder the planning sheet to match the order of the questions posed by the operating system. The planning chapter documents a fictitious network, nuts.com, that is used for the remainder of the book. The following six chapters are a network configuration tutorial based on the nuts.com network. Hunt then discusses what to do when TPC/IP doesn't work correctly at nuts.com. Then he discusses security at nuts.com. The book finishes with sources of additional information and a few selected reference sections.

The book is not implementation specific. Hunt concentrates on concepts, not release-specific step-by-step instructions. Your OS documentation gives you that information. Hunt compares the different ways UNIX systems are configured. That's important for those of us that are responsible for different flavors of UNIX. For example, chapter 5, the kernel configuration chapter, uses Solaris, Linux, and FreeBSD as examples. Unless you have the same OS version as outlined in the book, you can't configure your kernel with only the information Hunt gives. However, you'll understand the key concepts: Some UNIX systems like Solaris don't trust you to configure the kernel; some systems like Linux strongly encourage you to configure the kernel; a properly configured kernel is more efficient; different OS releases are configured differently. Hunt gives you an overview of kernel configuration without excessive details on configuring a system that you don't have.

Hunt, for the most part, includes everything you'll need to know about the subject without weighing the book down with trivia. I appreciate the unsaid trivia as much as the spoken butt-saving fact. We spend too much of our time reading information that serves no useful purpose. You don't always need to know everything about a subject -- just the important things. Hunt is selective with information. For example, when discussing firewalls, he gives a brief overview of firewalls without telling you how to build one. He ends the firewall discussion with the advice to hire it done unless you have time to kill. Too many authors describe the mountains that line the wrong-turn road.

The important things are usually said. However, I would have included two things that didn't make the final cut of the second edition: Santa Cruz Operation (SCO) specific information and configuring sendmail to occasionally use UUCP. The omission of UUCP is understandable. UUCP, while important to me personally, is not as important to the rest of the world as it was several years ago. The complete omission of SCO from the second edition, which was in the first edition, is not as understandable. SCO now owns the UNIX source code. This was not the case when the first edition was published. This isn't a serious flaw -- just a curious one. SCO systems are configured very much like the included systems. However, there are some differences. For example, the TCP/IP startup files are located in a different script. An SCO system starts TCP/IP in /etc/rc2.d/S85tcp while a Linux system starts it in /etc/rc.d/rc.inet1. Mentioning issues like these would have made a more complete book.

Regardless, Hunt's book is still one of the best UNIX sys admin books ever written. The second edition has all the strengths of the first. Moreover, it still fills a real need, though a different one. When the first edition hit the press, it described the obscure TCP/IP to a UUCP-dominated market. Hunt was an evangelist. Well, the world is converted and we need a tour guide. Hunt is now that tour guide. The second edition gives us a birds-eye-view of the TCP/IP terrain that can't be found in any one system's documentation.--Dr. Dobb's Electronic Review of Computer Books