Hardware problems are inevitable in networked computing environments. If you don't own tools that give you performance and error information, the chances are slim that you'll be able to diagnose such problems or prevent them from recurring.
The Windows 2000 Magazine Lab often performs tests that cause hardware problems. When we test a hardware product, we need to push the product to its limit to evaluate the device. However, stressing support hardware during software tests threatens test integrity. The Lab staff must ensure that stressed or poor-performing support hardware doesn't skew test results. Managed switches are invaluable tools for doing so, as well as for diagnosing and preventing network hardware problems.
During a recent software test, I borrowed a couple of Gigabit Ethernet network switches that offer out-of-the-box monitoring and management functionality. I installed these switches—a Summit48 Enterprise Desktop Switch from Extreme Networks and an HP Procurve Switch 2524 from Hewlett Packard (HP)—on my test network. The switches' integrated management interfaces and their various performance-monitoring and configuration options let me view traffic statistics about the support network I had built.
The switches' command-line and menu-driven textual interfaces let me use an RS232 connection for out-of-band management and Telnet for in-band management. Although the products' GUIs are somewhat slower than these text-based interfaces, the graphical access options are what truly make the devices easy to manage. Both products include an embedded Web server for easy graphical access to intuitive monitoring and management tools. The vendors also offer powerful management software for event logging, port naming, capacity planning, and multiple-device management tasks.
The GUIs gave me quick views of realtime port utilization, status, and error information. Each product's GUI broke down the port-utilization statistics into categories for granular network usage analyses. Both products listed statistics for unicast, nonunicast (e.g., multicast, broadcast), and error packets. In each product's GUI, I drilled down to the port level to view logically organized, detailed lists of packets that the port had transmitted and received. With minimal effort, I configured one of the switches to capture a specific port's performance data and save that data for later analysis. One quick look at this data helped me find systems whose connections were less than optimal (optimal being full-duplex transmission at the maximum connection speed).
The products' useful alerting capabilities required no additional configuration. One managed switch alerted me to a NIC that was causing excessive errors. I replaced the NIC, and the absence of subsequent alerts verified that this solution had eliminated the problem.
These managed switches proved indispensable in building and monitoring my test network. I could simply glance at one window to view port utilization and gather solid evidence that my network wasn't causing any bottlenecks during my testing. The switches' event-logging and alerting features let me forget my typical network worries and give my full attention to testing the software.
Although managed switches cost more than their unmanaged counterparts, prices have declined enough so that the cost difference is less appreciable than it used to be. At the same time, the cost of downtime that network problems cause. In a corporate environment, these switches won't take long to pay for themselves.