The art of network troubleshooting

Computerworld -

In many ways, the art of networking today is quite different from when I entered the field about 17 years ago as a part-time student assistant in the networking group at the university where I was working on my undergraduate degree. At that time, the network group had begun to dabble in a new technology called Ethernet, which promised to deliver data at the stunning speed of 10Mbit/sec. via coaxial cable (Thinnet, also known as 10Base2) between clients and servers.

While the technology has changed, the basic methods for troubleshooting networks really haven't. Sure, there are fancier sniffers, analyzers and monitors, but the real basics that demand an understanding of networking to the core level remain the same. Let me present a case history from the days of Thinnet to illustrate.

Lessons from history

Early in my career, while interviewing for a network engineering position, my potential supervisor and I engaged in a mutual story-telling session. I had recently solved a problem whereby users in a large building were experiencing severe network latency, and I felt sharing this experience in the interview would help demonstrate my networking skills.

The design was standard networking for the time; a Thinnet backbone running up one closet stack of the 10-story building, continuing along the ceiling of the top floor, and then back down the closet stack on the other side of the building. The backbone was connected to the campus network via a 10Mbit/sec. transceiver connected to a multimode fiber circuit that terminated at a Wellfleet router.

Latency only occurred when the network experienced a significant load, usually beginning shortly after 8 a.m. and ending at around 5 p.m. I plugged a Network General sniffer that was the size of a small suitcase and boasted a screaming 486 processor into a tap on the Thinnet backbone and observed multiple late collisions. This led me to conclude there was a physical layer problem.

The most basic method for testing the physical integrity of a Thinnet segment was to ensure proper cable termination. Thinnet required a 50-ohm resistor be placed on either end to prevent the signal from reflecting back along the cable by absorbing its energy. Removing the terminator at one end and using a digital multimeter to determine continuity and resistance was an intrusive but effective method of determining the backbone health. This test returned positive continuity and a resistance over 50 ohms, consistent with baseline observations of similar Thinnet segments yet on the high side for the estimated cable length.

Collisions were a normal functioning aspect of half-duplex Ethernet, but late collisions usually indicated either a station is not employing the CSMA/CD algorithm properly or a segment exceeded the maximum allowed length. The higher resistance reading pointed to the latter. Consultation with the network engineer who was responsible to this particular network design verified that recently the backbone was indeed extended at the furthest end to accommodate network connectivity for new offices.