Network Computingwww.networkcomputing.com

Bill: Which basically boils down to the "5-4-3" rule, with considerations for mixed media types.

Scott: The simple rule is that there should not be more than five segments, four repeaters and three populated segments among any DTEs in a single collision domain. A collision domain is defined as "a single Ethernet network in which there will be a collision if two DTEs attached to the system transmit at the same time."

Bill: A 10BASE-T hub containing DTE connections is a populated segment.

Scott: Therefore, we should play it safe by not allowing more than three cascaded and populated 10BASE-T hubs between any two DTEs.

Bill: Right, especially when mixi ng media types. Some examples of how to cascade hubs include using a cross-over cable between RJ-45 ports (some hubs have a built-in cross-over switch on one port for this function), a pair of 10BASE-F (fiber) micro transceivers, or a length of 10BASE-2 coax between two BNC connectors.

Scott: Many of the stackable hubs use proprietary interconnect schemes, so that the stack only counts as one repeater set. We often find ourselves, however, adding hub ports by simply connecting a cross-over cable to a second hub, effectively adding a second repeater set to the 5-4-3 equation.

Bill: Let's loo k at a couple of problem scenarios we've encountered in the field. We should note that in both cases, twisted-pair segments were within the maximum allowable distance of 100 meters. The fiber lengths were also within their spec.

Scott: In the first scenario, our customer started with a centralized fiber hub that connected to remote workgroup hubs.

Bill: The remote hubs were connected back to the fiber hub via AUI to fiber micro transceivers.

Scott: Most of the remote hubs were cascaded to one additional hub using a cross-over cable.

Bill: Looking at the worst-case path among nodes connected to the cascaded hubs, we quickly realized that there were four populated segments among stations.

Scott: One of those nodes happened to be a busy router.

Bill: S ome of the busier workstations were using this router, for which there were four intervening populated hubs.

Scott: As you can probably guess, the collision rates were high and fragments were appearing.

Bill: The short-term solution was to move the router from a remote hub port to a fiber port off the central hub.

Scott: Having done so, the collision counts and fragments dropped dramatically, since the "heavy hitters" were now within three populated segments of each other.

Bill: Over the long term, the remote cascaded hubs should be replaced with stackable versions or ones with higher port density, but the short-term solution of repositioning the router solved 95 percent of the problem.

Scott: The second scenario involved a simple topology where four hubs were connected in succession via fibe r transceivers or cross-over cables. Unfortunately, all hubs were populated with DTEs, giving us a total of four repeated segments.

Bill: At one end was a router, with the other hubs connecting to workstations.

Scott: Again, we noticed a high collision and fragmentation rate.

Bill: Our customer chose to solve this problem by adding an Ethernet switch in front of the final two hubs, which broke up a single collision domain into three--one from the router to the first two hubs in succession to the switch, and two new domains on the other side of the switch.

Scott: Remember that the 5-4-3 is only a simplified wiring model. For details on more complex wiring schemes and calculating the round-trip delay bit budget, we urge you to carefully consult the 802.3 standard.

Bill: Should we leave our reade rs with a caveat about some of those protocol analyzers out there that don't count collisions accurately?

Scott: Ah, yes. Low-end analyzers that use off-the-shelf workstation adapters, and even high-end analyzers based on PC Card adapters, have a bit of a problem counting collisions.

Bill: They simply equate fragments to collisions, not counting any of those collisions that appear in the preamble!

Scott: Nothing beats having a high-end cable tester (or an analyzer with specialized hardware) to accurately count those collisions and fragments.

Bill and Scott can be reached at otw@pmg.com. Portions of trace files from selected columns are available via Pine Mountain Group's Home Page (www.pmg.com).