Handling the Distributed Network

Strategies for Network Management

Issues in Managing Today's Networks

Network management is no longer a subcategory of systems administration; it's a job in itself. This new column will address evolving solutions for those who take on the challenge.

By Eddie Rabinovitch

Network management may mean different things to different people, based on their backgrounds and areas of expertise. To provide an overall framework, the International Standards Organization (ISO) defines network management as a combination of five functions:

Over time, this column will focus on each of these functions. It will also address other technical and management aspects of network management, such as definition and management of the service level agreement (SLA); asset management; selective outsourcing versus in-house network management and help desk; and tools and standards.

With the rapid growth and technical advance of communications technology, the task of network management becomes increasingly challenging. Due to the demand for powerful online applications and expanding data processing networks, the complexity of network topology and communications equipment becomes more sophisticated. Although each individual network element may be well-understood, combining the elements into a complete network--and then understanding the interrelationships and end-to-end effects of these combinations--is a formidable task for which there are no easy answers or quick references.

Often the new networks were put in place according to the Nike slogan: Just Do It. In these cases network management is left for a later time. Because network managers are usually busy fixing existing problems, they don't have much time to look ahead and develop procedures for proactive network management. Therefore, management of distributed networks in client/server environments remains a to-do item on many network managers' agendas. While the number of client/server applications continues to increase, and as more power and intelligence are deployed in remote locations, these managers face a battery of new challenges before they have solved the old ones.

Let's consider one common example. Antiquated but still common, hierarchical master/slave networks usually have simple topologies, and there are established techniques and tools to address performance management of such networks. This task becomes much more difficult in dynamic client/server networks, whose network topologies change frequently. Therefore, it is usually difficult to estimate and predict the exact route taken by a particular message. And modeling and simulating such networks also are significantly more difficult than in the relatively simple hierarchical networks.

SNMP Rules

Nevertheless, there are solutions to these problems--often more than one. Only a few years ago, most industry analysts were placing their bets on the Common Management Information Protocol (CMIP), a part of the Open Systems Interconnect (OSI) standards developed by the Europe-based ISO. Today, the Simple Network Management Protocol (SNMP) has emerged instead as the de facto industry standard. This occurred for much the same reason as the TCP/IP protocols won out over OSI as a whole: Many people and products were using SNMP and did not want to wait until CMIP is finalized and productized to go forward with their architectures.

SNMP became the standard of choice for network management of Unix systems, LANs and distributed WANs. For example, about five years ago IBM planned to incorporate CMIP into its NetView network management suite. Today IBM has an SNMP-based network management strategy.

SNMP defines a set of standard Management Information Base (MIB) variables that include vital management information, such as make, model and serial number of the device; error counts and statistics; on/off status of a component; and others. To address SNMP's shortcomings (which include overhead and performance issues, and security exposures), the Internet Engineering Task Force (IETF) initiated development of SNMP version 2, which is still under development, mainly because of the differences in flavors proposed by competing vendors and some political issues within the SNMP v2 camps.

Because the vast majority of network management tools and techniques on the market today is based on SNMP standards, this column will focus primarily on how to manage today's network using SNMP.

More Art than Science

While SNMP provides a solid standards base for distributed network management, using the current generation of network management tools requires more creative skills than scientific ones. Not many firms can afford to keep skilled technical personnel in their remote branches; the trick is figuring out how to serve and manage the equipment that is distributed throughout the network, while employing a limited number of people and a minimal number of central sites. Such an approach requires a centralized focal point of control, data distribution, backup and security.

Yet management of distributed networks currently suffers from a lack of integration capability. While vendors such as Cabletron, Hewlett-Packard, IBM and Sun Microsystems are developing strategic architectures and deploying the first modules of distributed network management systems, most networks are managed with a collection of loosely integrated products.

There are three different levels of distributed systems management integration. The most popular (and simplest) approach is integration at the interface level, where different component of the systems are integrated in a network management workstation. With this approach, integration of different components relies on a single administrator to manage them. However, since the applications do not share information and the events are not consolidated across different management applications, its usefulness is limited.

The next level of integration is at the event level. With this approach, error conditions (traps, events and alerts) are consolidated between the network management platform and systems management applications. Some tools offer integration at the SNMP level, where applications update an MIB extension with status and trap information. This information flows through the SNMP hierarchy to the manager, where events are consolidated from multiple sources. This is a strictly SNMP standard approach, not based on any proprietary software or hardware components.

The highest level of integration is the object level, where network management applications communicate with other applications by passing messages. For example, a network topology object can service all distributed systems management applications requesting the status of networks and nodes prior to performing the required functions (such as backup or software distribution). Most vendors are planning to provide this level of integration, but only a few offer real products in this area today.

Most network managers follow the simple rule of better safe than sorry. The result of this caution often is over-engineered (and overpriced) network infrastructures. To address this phenomenon not only reliable products but sound procedures and processes for network management are required. Early signs are here of cross-platform integration of network management products for client/server environments on mainframes, Unix systems and LANs. Future inistallments of this column will expand this discussion and try to provide specific recommendations on issues and possible solutions.

Eddie Rabinovitch is senior manager with the network and desktop consulting practice at Unisys Corp. in Blue Bell, PA. He can be reached at EddieRabinovitch@unn.unisys.com.