Enterprise Network Management In Perspective
Don't buy a network management application! At least don't buy one now. We spent months testing on a 10,000-node network, and these products are just not ready. Out-of-the-box status, the stability and performance to handle 10,000 nodes, and improved packaging are beginning to show up, but none of the products has it all.

We saw network status innovations in Tivoli Systems' Netview, Loran's Kinnetics and, to a lesser degree, Hewlett-Packard's Network Node Manager, and even though network services like DNS and DHCP are being monitored, no one is monitoring real applications. Cabletron's Spectrum and Bull's OpenMaster still require significant implementation work to get network status.

Web implementations are getting better in some of the products, but overall are still lacking. Here again, Netview, Kinnetics and Network Node Manager are turning the corner. Most of these products can manage more nodes than their previous incarnations could, but instability and lack of responsiveness under that increased load cancel out the improvements. The struggles were most notable in the beta versions of Netview and Network Node Manager, with OpenMaster being just plain slow.

Perhaps the most encouraging news involves pricing. Bundling has reduced, but not eliminated, the need for add-on modules to handle basic SNMP management. The flattening of price structures and improvements in NT have resulted in in lower initial costs for most of the products.

Our recommendation: Avoid purchasing a network management product until the shortcomings are resolved. How long will it be? It's our opinion that Netview and Kinnetics have the best chance of becoming viable within the next sixth months.

How We Tested Network Management
To differentiate these enterprise network management products, we focused on the status, deployment, performance, inventory, learning curve and reporting that each product provided. These features and implementations are the ingredients that comprise a useful network management solution.

Testing was performed in Network Computing's Syracuse University lab, where a combination of the university's live network and our own corporate network offered the ability to load each server with more than 10,000 SNMP and ICMP (Internet Control Message Protocol) nodes. This kind of loading stresses not only the hardware, but the management software's ability to deal with the ever-changing status of so many nodes as well. While the number of nodes we used is excessive, we wanted to find out what loads were possible. In the past we've tested a number of Windows NT-based network management products and found that they performed poorly (see "Can NT Balance the Network Management Load?" at www.networkcomputing.com/821/821f1.html).

Network status monitoring should be the primary function of an enterprise network management application. We were looking for monitoring that went beyond simple SNMP gets. We wanted to see application and service monitoring, such as Mail and DNS, and we wanted to know about calculated statistics, such as availability and utilization. We were also on the watch for the ability to modify these calculations, the relative ease of use of the interface, and what you as a user need to know about the architecture of the application.

In addition to status monitoring, we evaluated network mapping. Accuracy, of course, is crucial, but even more important are the tools that manipulate and clarify the view of the network. Therefore, reports, filters, sorts and discovery controls were as critical as accuracy of the initial map display. We ran our discovery in a stepwise progression using a seed router, and then expanded each subnet to give each product the best possible chance at accuracy.