Production-quality enterprise networks The secret to networking success is to plan for change

February  1996

Plan carefully and put the kinds of controls in place that will ensure an enterprise network's success from the outset. Four main areas need equal attention when implementing and managing an enterprise network: network complexity and change, personnel, scope, and network management. (1,400 words)

Mail this article to a friend

It's obvious you need some sort of network for a distributed-computing environment -- it's the conduit through which clients and servers communicate queries, processes, and data. Our view of the network, however, transcends the common and simplistic pipeline metaphor; for us, The network is the computer. A distributed network for your New Enterprise is more than just the sum of wires, interfaces, and protocols. We elevate the collection of simple networks with desktops encircling departmental servers up to a unified system of many servers that provides users with global access to information, communications, and a variety of client/server (often mission-critical) resources. To do this, we build a world-class network infrastructure and manage that New Enterprise network with world-class processes and tools. In other words, we build what we call a "production-quality" enterprise network.

Five characteristics of an enterprise network
The importance of the network in the New Enterprise cannot be stressed enough. Its qualities and potentials are derived from five important characteristics that also distinguish the enterprise network from its many peers and predecessors.

1. The enterprise network should be a competitive advantage to your entire organization. It connects all computers in the enterprise and gives users access to information and computing services across the entire organization. This means there is networking support over a common backbone for a variety of protocols, operating systems, and applications.

2. An enterprise network is integrated. Some networks give users access only to specific applications. For example, some companies have two networks: one for access to legacy mainframe applications and another for access to desktop productivity tools and shared services. What good are those productivity tools when all the important corporate information is buried in the mainframe data warehouse? By definition, our enterprise network gives users reasonable access to all data and applications, including emerging applications that use new technologies.

3. An enterprise network is global. It must have an external view beyond the physical and political boundaries of the organization. There is far more and important information about your customers and competitors available on global systems than what your internal network can contain or manage. The New Enterprise gives its users access to applications, data, and communications internationally.

4. An enterprise network is adaptable and extensible. It can support current requirements as well as future enhancements without major modifications. It can change as user requirements and technology change. For example, we expect future enterprise networks to support more than simple data models to include multimedia applications without major renovations.

5. An enterprise network has production qualities. This simply means it is manageable and has high levels of RAS (reliability, availability, and serviceability) to support our mission-critical applications and other production systems. This remains one of our major themes: how we implement networks with production qualities.

Guidelines for success
Simple in concept and awesome in scope, the enterprise network is an intricate and complicated web of technology. It only makes sense that you should plan carefully and put the kinds of controls in place that will ensure its success from the outset. In our experience, there are four main areas that need equal attention when implementing and managing an enterprise network. Unless you properly deal with network complexity and change, invest in personnel, keep a broad scope, and carefully plan and manage your New Enterprise, all will surely transform into the four horsemen of the networking apocalypse.

There are an infinite number of ways to connect the hardware and software in a network. (In practice, only a few actually work -- we will detail our successful architecture in a future column.)

To add insult to injury, distributed-computing networks are highly integrated, so modifying one component often will effect the operation of others, no matter how apparently isolated they may be. That can lead to disaster in an undisciplined network, after all, change is the way of business in the New Enterprise. New technologies, new software releases, hardware upgrades, and user requirements can provide strategic as well as tactical advantages for adapting to today's fast-paced and competitive world. You can't avoid the problems and constraints that complexity and change bring to your enterprise network. You can, however, facilitate and manage them through a structured, standardized network architecture.

The New Enterprise also needs a well-trained staff capable of implementing and managing your network. Sure, training is time consuming, and it places an extra burden on limited resources, but that's penny-wise and pound-foolish thinking. We plan for adequate personnel training from the very beginning, and we use a variety of alternative curricula, such as computer-based training and online information services.

Make sure the plan and design of your enterprise network has a broad scope, preferably global. Although typically departmentalized for efficiency, organizations don't exist for long if parts of the business starve. Don't balkanize the network design, or it will surely become an impediment to change, and, worse, it can cost you plenty for interconnectivity and interoperability after the fact.

Last, but most important, plan to manage your network well. Put in place the disciplined tools and practices that will help manage your network. Don't wait for a disaster to happen before you plan for a cure. Invest in forethought and planning. Preventive medicine is less difficult and less expensive, plus the patient usually lives.

Cost of ownership
Distributed-computing environments are primarily built from nonproprietary, "off-the-shelf," and, therefore, inexpensive components and software. Yet, it is common wisdom that client/server, distributed computing is costly, even compared with traditional computing systems. Why? For many distributed-computing pioneers, it was the cost of owning a network that got out of control. For us, client/server distributed-computing technologies hold to their promise of better computing services for less cost. We couldn't have achieved those goals, however, if we hadn't implemented and managed the network properly. Again, planning is the key to controlling costs.

The total cost of owning a network includes equipment, operations, and downtime. While the list of networking equipment is long -- cables, connectors, patch panels, concentrators, routers, gateways, management systems, protocol analyzers, network adapters, cable testers, and so on -- we've found the cost for equipment is surprisingly small compared with network operations and downtime. Networks need constant monitoring and maintenance to keep them up and running smoothly. There are new users to connect and established users to relocate. And there is always a queue of new software releases and hardware upgrades to make. In the distributed-computing environment of the New Enterprise, all these modifications mean changes to the network. Each must be planned, tested, evaluated, implemented, and monitored. If we're lucky, we will not introduce too many new problems with these changes. Regardless of how endeared you are to lady luck, problems will occur in the network, and downtime will impact your organization. Like other things, the cost for network downtime -- potentially millions of dollars per hour -- depends on how well you manage your network. Control downtime and you control a major, if not the major cost for your enterprise network.

There's one more important point about network cost: You get what you pay for. While equipment is only a small part of the total, poor equipment will drive up the more burdensome costs for operations and downtime. If you expect a distributed-computing network to support mission-critical applications, you need to purchase reliable equipment, even if it is more expensive in the short run.

A multidimensional approach
As we continue to chant the mantra through our columns and books, we apply the same RAS disciplines to our distributed network and computing systems as have been successfully applied for managing mainframe systems. But are they really sufficient to fully manage the New Enterprise? Almost. Actually, we use a multidimensional approach that combines the hard-nosed disciplines of mainframe controls with the flexibility and responsiveness that are hallmarks in this age of distributed computing. There's a balance to be struck.

On the one hand, the New Enterprise has evolved because businesses demand that users get the mission-critical services they need when and where they need them. In today's successful enterprise, therefore, users dictate their computing needs. The job of corporate IT, on the other hand, is to deliver computing services, making sure those critical systems and technologies are in place and running smoothly when users need them. You can't do that job unless you have disciplined control over the systems. To strike the balance, get users actively involved in their network developing plans, defining requirements, and implementing systems. And be flexible: User requirements and technologies change frequently and are sometimes unpredictable. We do not avoid change, but recognize its importance in the New Enterprise. Plan for and manage change.

Resources

• Other Unix Enterprise
  /sunworldonline/common/swol-backissues-columns.html#unix

If you have technical problems with this magazine, contact webmaster@sunworld.com

URL: http://www.sunworld.com/swol-02-1996/swol-02-unix.html
Last modified: