Ethernet Switch Tests Improve Network Reliability

Components Make a Difference
Problems can arise from variations within an electronic component’s specification. Those variations produce components with different characteristics, which produce switches with varying characteristics. Hand-assembled devices such as transformers can be susceptible to enormous skews in tolerance from one lot to the next. If a front-end transformer used in a switch’s Ethernet port was manufactured out of spec (manufacturers still hand-wind these transformers), it might produce total product failure or a reduction of Ethernet cable length that a port can drive. Loopback tests typically miss this problem.

In addition, a capacitor with dielectric characteristics suitable for filtering at low frequencies may not filter well at higher clock frequencies. Because of these capacitors’ availability, design engineers may select a capacitor that is inappropriate for a circuit. As the industry produces higher network speed designs (e.g., gigabit Ethernet), network components must meet more stringent tolerances for clock and other timing attributes. In gigabit Ethernet, the shortening of signal risetimes that accompanies an increase in clock speed can lead to improper timing in network signals. The timing errors increase the number of error packets, which severely limits the performance of an Ethernet switch or other network component.

Small revisions in board layout can affect a switch’s network performance. At low frequencies, a minor change in a PCB’s stub length may have minimal effect on network performance. At higher frequencies (100 MHz and up), small layout changes can make a difference in how well the switch performs. If a design change calls for trace changes because of EMI emissions or susceptibility, side effects such as cross talk, improper signal timings, and other unpredictable or difficult-to-test effects may occur. Ultimately, small design changes can cause an Ethernet network component to deviate from the IEEE standards for Ethernet.

Emulate a Network
Instead of performing a quick functional test, you should test a network component by emulating a live network. Using a test configuration like that shown in Figure 1, you can inject random “real world” network problems such as collisions and fragmented packets into the test environment from a single-ended transmitter. That will expose some of the problems mentioned above.

Figure 1. A single-ended transmitter can inject timing and packet errors into a simulated network.

The test configuration links external hubs to packet-generation nodes, creating a network-like segment to feed the switch under test. You should set the remote nodes in the simulated network to echo mode and have them return one or more packets for each packet they receive. You will then generate a client-server stream of “randomized” traffic.

You should test Ethernet switches under various loads and verify remote monitoring (RMON) counters and Simple Network Management Protocol (SNMP) Management Information Bases (MIBS). In addition, you should perform power-up tests as another real-world example. Power-up tests can tell you if the DUT accurately and completely executes auto-negotiation. Auto-negotiation occurs if you stress the switch and it adapts to all advertised rates and duplex modes. The test also tells you how well the switch performs on power-up under a heavy traffic load.

Parametric tests will reveal defects that packet-level tests may not. Timing inconsistencies can vary slightly among tranceiver ICs used to drive Ethernet ports in network components. These variations can cause interoperability problems between transceivers from different IC manufacturers. Those problems can affect a switch’s ability to communicate consistently to data-communication equipment from different manufacturers.

You can’t diagnose the defects I’ve mentioned with higher-layer testing. These defects require a physical layer test such as one that examines signal characteristics—risetime, amplitude, jitter, and cross talk. The real value of manufacturing testing lies not just in filtering out bad products, but in feeding important defect information back to the design engineers. T&MW

FOOTNOTE
1. IEEE 802.3, 1998, Information technology - Telecommunications and information exchange between systems—Local and metropolitan area networks—Specific
requirements—Part 3: Carrier sense multiple access with collision detection (CSMA/CD) access method and physical layer specifications. IEEE, Piscataway, NJ. www.ieee.org.

Vinod Bhardwaj is the founder of several companies including ControlNet and Antara. Prior to starting these companies, he founded Kaplana, where he designed the first Ethernet switches. vinod@controlnet.com