Stations are connected in a physical or logical RING. A small control frame — the TOKEN — circulates continuously. A station may transmit ONLY while it holds the token. THE TRANSMIT CYCLE: 1. a station wanting to send WAITS for the free token 2. it CAPTURES the token by flipping one bit (the token bit) in the frame header, converting the token into the start of a data frame 3. it appends its data and transmits 4. the frame travels around the ring; every station forwards it, and the DESTINATION copies it while also forwarding 5. the frame returns to the SENDER, which removes it from the ring ("drains" it) 6. the sender releases a NEW free token Note step 5: the frame makes a complete circuit. The sender is therefore the one that removes it, which conveniently provides an implicit acknowledgement — if the frame comes back with the address-recognised bit set, the destination existed and saw it. WHY NO COLLISIONS ARE POSSIBLE: exactly one token exists, and only its holder transmits. This is not collision *detection* (Ethernet) or collision *avoidance* (Wi-Fi) — it is collision *impossibility* by construction. THE 802.5 FRAME: SD AC FC DA SA DATA FCS ED FS │ │ │ │ │ │ │ └ frame status │ │ │ │ │ │ └── ending delimiter │ │ │ │ │ └────── frame check sequence │ │ │ │ └──────────────── source address │ │ │ └──────────────────── destination address │ │ └──────────────────────── frame control │ └──────────────────────────── access control └──────────────────────────────── starting delimiter The ACCESS CONTROL byte carries: PPP T M RRR │ │ │ └── reservation bits (request a priority) │ │ └───── monitor bit (loop detection) │ └──────── token bit: 0 = free token, 1 = data frame └───────────── priority bits (0–7) So the difference between "a token" and "a data frame" is one bit. That is why capturing the token is cheap — no new frame needs to be generated.

PRIORITY OPERATION — the feature Ethernet had no answer to: · the token carries a current PRIORITY (0–7) in the PPP bits · a station may transmit only if its frame's priority ≥ the token's priority · a station wanting higher priority writes its request into the RESERVATION bits as the frame passes · the station that releases the token raises the token's priority to the highest reservation seen, and REMEMBERS the old value so it can lower it again later → real quality of service in 1985. Voice or control traffic could be given guaranteed precedence, which is exactly what Ethernet needed VLANs and 802.1p to approximate a decade later. THE ACTIVE MONITOR — one station is elected to police the ring: · removes ORPHANED frames — a frame whose sender has died and which would otherwise circulate forever. It uses the MONITOR BIT: the monitor sets it to 1 on first pass; if it sees a frame already carrying 1, that frame has been round twice and is drained. · regenerates a LOST token if none is seen within a timeout · provides the master clock for the ring · detects and removes duplicate tokens STANDBY MONITORS watch the active monitor and elect a new one if it fails. THE COST OF THIS DESIGN: the ring has a single point of coordination and — physically — a single point of failure. One broken cable or dead NIC breaks the ring for everyone. IBM's answer was the MAU (Multistation Access Unit), a wiring concentrator that bypasses a failed station automatically, which made the physical topology a star while the logical topology stayed a ring. That is the same physical-star idea Ethernet hubs used, arrived at from the opposite direction.

TOKEN RING (802.5) ETHERNET (802.3) access method token passing CSMA/CD contention collisions IMPOSSIBLE possible, detected determinism YES — bounded wait NO — statistical behaviour under DEGRADES GRACEFULLY COLLAPSES: more heavy load (throughput holds) collisions → more backoff → less throughput behaviour under WORSE — token BETTER — a station light load passing overhead transmits at once even with one station wanting to send priority support YES (8 levels) no (until 802.1p) single point of yes (ring/monitor) no (with a hub or failure switch) speed (historical) 4 / 16 Mbps 10 / 100 / 1000+ cost per port HIGH (proprietary) LOW (commodity) vendor IBM-dominated multi-vendor WHY ETHERNET WON, in order of importance: 1. COST. Commodity NICs and cabling from many vendors versus IBM's proprietary MAUs. Price beat elegance. 2. SWITCHING. Once Ethernet switches replaced hubs (mid-1990s), each port became its own collision domain — and Ethernet's one real weakness, collisions, VANISHED. Token Ring's central advantage evaporated overnight. 3. SPEED. Ethernet went 10 → 100 → 1000 Mbps while Token Ring stalled at 16. 4. SIMPLICITY. No monitor election, no token maintenance, no ring-break recovery. THE DECISIVE POINT is (2). Token Ring's determinism was worth paying for only while Ethernet had collisions. Switched Ethernet is ALSO collision-free, at lower cost, so the trade-off that justified Token Ring simply stopped existing. WHERE DETERMINISM STILL WINS: industrial and safety-critical networks, where a bounded worst-case delay is a requirement rather than a preference — the same hard-real-time argument as in ACtE0705. PROFIBUS, and modern Time-Sensitive Networking (TSN, IEEE 802.1Q extensions) bring scheduled, deterministic transmission back to Ethernet itself, which is the idea Token Ring had, implemented on the hardware that won.