Token Ring Reliability

1. Reliability is a much bigger issue for token rings than it is for Ethernet.

   • A token ring has many active components that could act a "single points of failure".

   • There are several ways in which "transient" problems could lead to permanent problems.

2. The basis of the token ring's mechanisms from recovering from "transient" errors is the assignment of the role of "monitor" to one of the ring's stations.

   • For example, one of the monitor's jobs is to timeout if it does not see a token or packet for longer than the maximum transmission time plus the ring's propagation time.

   • This job can't safely be shared by all stations since if two stations timed-out nearly together a ring might end up having several free tokens.

3. To ensure reliability even if the monitor fails, the assignment of the monitor task to a station is not fixed. When necessary (for example when the machines on the network are first turned on or when the previous monitor crashes or is turned off), the stations of the ring "hold an election" to choose a monitor.

   • When a station realized that a new monitor needs to be elected (we will discuss how it does this later), it sends out special "election" packets (the text calls it a claim token) containing its own hardware address.

   • While sending these packets, it listens for incoming packets.

     • If it receives a packet from another station, it compares its own address to the address in the packet.

       • If its address is smaller, it stops sending packets of its own and instead forwards what it receives.

       • If its address is larger, it just gobbles up the incoming packet and keeps sending its own.

     • If it receives one of its own election packets back, it claims victory and starts acting like a monitor:

       • First it sends out a message to make sure that all the other stations know that the election is over.

       • When this packet returns the new monitor sends out a free token.

4. Of course, given that the monitor is very important to the rings ability to handle failures, we need a way to handle the failure of the monitor itself.

   This can be handled by having station's timeout if they don't see a free token for a very long time.

   Alternately:

   • Each packet contains a bit called the monitor bit.

   • When a station sends a message it stores zero in the monitor bit.

   • When a message passes through the monitor, the monitor sents the packet's monitor bit to 1.

   • If a station ever see one of its own packet's return without having the monitor bit sent, it suspect trouble and starts an election for a new monitor.

5. The monitor bit also solves another problem. If the monitor ever sees a packet with the monitor bit set, it will realize that the packet has been around the ring at least twice. Whichever station was suppposed to remove the packet from the ring has failed! In this case, the monitor removes the packet from the ring and sends a free token. What order should the monitor bit and the free token bit fall to make this possible? Why aren't they in this order according to the readings?