1.3Physical interfaces, cabling, and switching

Sites often report "for some reason it will not reach the next building" or "the switch CPU spikes and traffic slows." Many such issues stem from wrong cable selection or a shaky grasp of basic switching behavior. This section covers the judgment of choosing cable by distance and speed, and how an L2 switch learns destinations and handles frames it has not learned, in a form that connects directly to fault diagnosis.

1.3.1Cabling and physical interfaces

• Copper (UTP twisted pair)—Cat5e/Cat6/Cat6a, etc. Cheap and easy to route, but in Ethernet the rule is up to 100 m per run, and it is susceptible to electromagnetic noise. It is used for horizontal cabling within a floor (to PCs and APs).
• Optical fiber transmits with light rather than electricity, resists noise, and suits long distance and high speed. Single-mode (SMF) has a thin core, uses a laser, and serves long-haul backbone spans of several to tens of km; multi-mode (MMF) has a thicker core, uses an LED/VCSEL, and serves mid-range spans within a building up to a few hundred meters—choose by the cost/distance balance. SMF reaches farther.
• Copper wiring depends on use: straight-through between unlike devices (PC<->switch, switch<->router) and crossover between like devices (switch<->switch, PC<->PC) is the classic rule. However, modern switches with Auto-MDIX automatically detect and absorb the wiring difference, so a link often comes up even if the cable type is mismatched.

1.3.2Switching behavior

• An L2 switch records the mapping of a received frame's source MAC address to its receiving port in the MAC address table (learning). Thereafter, frames for that destination MAC are forwarded only out the matching port. In IOS you can view it with show mac address-table.
• If the destination MAC is unlearned, or the frame is a broadcast/unknown unicast, the switch performs flooding, sending it out every port except the one it arrived on. Learned entries have a lifetime (the aging timer, Cisco default 300 seconds); an entry idle for that period is removed and re-learned.

Suppose you are assigned to merge the LANs of two buildings about 800 m apart into one. Trying to connect them with UTP (copper) "because it is cheap" is a wrong call—Ethernet UTP is limited to 100 m per run, so 800 m simply will not reach, and outdoor copper also risks lightning surges and noise. The right answer here is optical fiber, and for 800 m with future higher speeds in mind, choosing long-distance-capable single-mode (SMF) is sound (within one building over tens of meters, cheaper MMF suffices, but for an 800 m inter-building span SMF is the natural choice). After the merge, users in one building report "traffic is occasionally slow and the switch load is high." Running show mac address-table shows the table holding oddly few entries, with many frames being flooded. Tracing the cause reveals MAC address table instability—an L2 loop destabilizing learning so the MAC table flaps—and the resulting flood of unknown unicast was consuming bandwidth and CPU. The lesson: if you understand the switch's basic behavior that "an unlearned destination leaks to every port (flooding)," you can go from symptom (load, latency) to root cause (learning breakdown) by the shortest path. Cable selection follows from distance and speed; switching faults follow from the learning/flooding principle—in both, understanding the mechanism translates directly into diagnostic speed.

1.3.3Section summary

• UTP reaches up to ~100 m; beyond that, or in noisy environments, use optical fiber—SMF for long haul, MMF for mid range
• An L2 switch learns the source MAC + receiving port (show mac address-table), floods unlearned/broadcast frames, and ages entries after a default 300 seconds
• Modern switches use Auto-MDIX to auto-absorb straight/crossover differences. For load/latency symptoms, tracing the cause via the flooding principle is fast