High Speed LANs

1. There are two basic measures of the performance of a medium access control protocol:

   Delay

   measures the difference between the amount of time required to deliver a packet using a given access control protocol and the time required if the sending station had exclusive access to the network.

   Efficiency

   (or utilization) measures the amount of data delivered per unit time compared to the amount that could be delivered in the same time if medium access control was not an issue.

2. I have already pointed out that the ratio of the time it takes a packet to propagate through a network to the time it takes to transmit a packet has a significant effect on the efficiency of most protocols.

   • For unclear reasons, this ratio is frequently called "a".

   • The value of a can be thought of as the number of average packets required to just fill the network with bits. In fact, it should be noted that "a" becomes large when either the transmission rate increases, the packet size decreases or the network size increases.

   • At high load, the main factor producing inefficiency in a token ring is the time the token spends moving from one station to another. If this time is measured in bits rather than microseconds, then the larger the value of "a", the more bit transmission times get wasted as the token moves from one station to another.

     • In the paper by Stallings I included in the readings, he gives a bound on the efficiency of token ring.

     • Basically, assuming that a ring contains N stations, that the packet size used in the calculation of "a" is the size of the maximal number of bits a station on the ring can send before passing the token, that all stations have data to send and that the time to send such a packet is one unit, then the time spent sending a packet is 1 plus the time it takes the token to move between two stations on average, a/N. So, the efficiency is at most

       (1)/(1+a/N)

     • At low load, the average delay introduced waiting for a token to arrive is half the propagation time of the ring regardless of the transmission rate. As the transmission rate increases, however, the delay increases as a fraction of the total time required to transmit and deliver a packet.

   • As "a" gets larger, the efficiency of contention protocols such as CSMA/CD also suffers.

     • Basically, if we again assume our units of time are such that the average packet transmission time is 1, then the time wasted by each collision is proportional to the x time required for a to propagate throughtout the network, a. If "C" is the expected number of collisions, then the time to successfully send a single packet will be 1 + k a C for some constant k. Therefore, the efficiency of the protocol will behaves as

       (1)/(1+k a c)

     • One would expect the "C" in this formula to increase as the number of stations increased. Thus, unlike token ring, CSMA/CD gets worse as either "a" or the number of stations increases.

3. The problems associated with large "a" values are typically caused by increases in transmission speed. Therefore, LANs in which "a" is large are considered High Speed LANs even though a large "a" value may actually be the result of an increase in network size.

4. A great deal of effort has been devoted to the design of protocols that operate efficiently on high speed LANs (a.k.a. HSLNs).

   • A fair amount of work has sought to produce bus networks that could compete with ring networks.

     • The reliability advantages of a passive bus helped motivate this work.

   • Most of this work has assumed that the physical layer of the networks involved would be based on optical fiber. As a result, many of the designs are based on buses with uni-directional data flow.

5. We will first discuss collision free protocols for uni-directional bus networks. Then we will look at two alternate approaches to the design of ring protocols: slotted rings and insertion rings. Finally, we will look at a slotted bus protocols, DBDQ.