Make your own free website on
CMPnet Click here to visit UUNET

 Site Guide
 Latest Updates
 '96-'98 Articles Index

 Search NWC:
 best match

 Technology Guides
 Wide Area Systems
 Network Management
 Collaborative Computing
 OSes & Services
 Servers & Peripherals

 On Our Site
 Network Design Manual
 Interactive Buyer's Guide
 Interactive Report Card
 Online News
 Real-World Labs
 E-Mail Poll
 Frezza's Forum
 Workgroup Computing
 ITpro Downloads
 Well-Connected Awards

Sponsored by:
Click here for Sprint

 CMPnet 3-D Site Map

 Site Services
 Advisory Forum
 Network Computing Links
 Best Jobs USA
and Marketing

Visionary Articles

 Web Connection
 Industry White Papers
 Special Supplements

CMPnet Resources
 Site Map
 Ad Info

 Free E-mail
 Sign Up Now 

Get PointCast - free!

Designing-And Redesigning-Today's Local Area Network

by Art Wittmann


You've built a network and it runs reliably! Congratulations; you're doing better than many. But you know - deep in the pit of your stomach - that sooner or later it's going to be time to rebuild your network.

Reasons to Rebuild Your Network

There are three big reasons to build a new network and a couple of lesser ones. The three biggies are:

  • A substantial increase in the number of users on your network.
  • A substantial change in the power of the average user's workstation
  • New applications emerge that demand more or different network services.

Lesser reasons include:

  • Making the network more manageable for changes, moves and adds.
  • Adding redundancy and improving reliability of the network
  • Updating out-of-date equipment

Note that the big reasons are very much user driven, while the lesser reasons are mostly driven by the needs of the network manager - that's you. It isn't that your needs are inconsequential, however, satisfied users are what turns the budget wheels, so they get first consideration.

Design Philosophy: Switch When You Can, Route When You Must

Until recently, routers were the only game in town for adding bandwidth to networks. Some larger networks have been built with transparent bridges , however, these networks usually proved to be difficult to scale and manage.

Switching hubs have become much more popular over the last few years and now offer the features necessary to build a large, reliable high-performance network. Switching hubs initially where nothing more than multiport bridges, offering little more than bandwidth. Now, with virtual LANs and some layer-three protocol processing, switching hubs can be used to safely build economical high-performance networks.

In this article we will take a fairly progressive view of routers and switches. Our philosophy throughout will be to switch where you can and route where you must. Some vendor has probably already coined the phrase, but it is a good catch-all for the advice that we will provide throughout this document.

Our reasoning here is simple:

1. Routers are software-driven devices that excel in flexibility and feature sets. Generally, much, if not all of the routing decisions are determined by algorithms run on general purpose RISC CPUs. Because of this, routers are:

a. Expensive on a per-port basis. CPUs and memory cost a lot of money and router vendors extract heavy margins to support their ongoing software development efforts.

b. Routers are not particularly fast. CPU algorithms take time to run and, given the chance, you and I usually load up routers with all kinds of rules and control lists that must be checked on a per-packet basis, slowing the router even more.

c. Routers are a great way to get from a trusted network segment to an untrusted segment. Generally traffic between such segments (say, between an engineering department and a marketing department) is orders of magnitude less than within a department. Also, for all of the reasons we gave for routers being slow above, they also make great firewalls.

Switches, on the other hand, are firmware-driven devices. Virtually all of what they do has been committed to silicon in the form of Application Specific Integrated Circuits (ASICs). Custom ASICs can provide lightening fast algorithmic processing, but they allow for fairly little flexibility in the algorithm run. As a result switches are:

a. Simple. They take in packets, find a path for them and spit them back out another port. Network managers can't set up a large number of parameters on a per-switch basis,

b. Cheap. Particularly Ethernet switches have been reduced down to a few chips and usually only one major chip per port. They are beginning to rival the price of non-switched intelligent hubs.

c. Effective. Because they are simple devices, they can deliver exactly what you want - bandwidth to users. New technology like virtual LANs (VLANs) and the ability to deliver VLAN traffic outside of the box make the manageable on most all networks.

Another assumption we'll make is that you need more performance and flexibility out of your network. We'll also look closely at some management issues. These are three key areas that matter to you as a network admin. and that vendors use to distinguish themselves.

Table of Contents

November 15, 1996

Print This Page

e-mail E-mail this URL

NWC Home Site Guide WANsites UnixWorld Network Design Manual
Careers Network Computing Links ITpro Downloads Interactive Buyer's Guide
Online News Latest Updates Real World Labs Subscriptions WorkGroup Computing


CMPnet Click here to visit Adaptec!