previous | contents | next

418 Part 2 Regions of Computer Space
Section 5 ½
Networks

fact the original motivation for constructing the ALOHANET, while the two-channel configuration was primarily chosen to allow this investigation without complication from the relatively dense total traffic stream being returned to all users. An additional reason for the star configuration was the desire to centralize as many communication functions as possible at the MENEHUNE, minimizing the cost of the TCU at each user node.

Within this context, a number of protocol issues must he resolved. The more important of these are:

Many of the original choices in these areas have undergone significant changes as a result of new user resources and user interfaces, or in some instances due to advancements in theoretical knowledge. The addition of repeaters has (potentially) a particularly significant impact on protocol.

We now discuss some of the considerations and resulting choices made in each of the above areas, with the impacts of new factors introduced within the context of each area. The section concludes with a brief discussion of the problem of integrating file traffic into the random access channel, a subject of current concern in the ALOHANET.

Random Access Channel Control

The retransmission strategy used in the random access scheme plays a central role in the scheme's effectiveness. Its determination directly affects the average delay experienced by users for a successful transmission, given a certain number of users accessing the channel, their traffic statistics, and the channel capacity. It can also be used to prevent the occurrence of channel saturation, a situation in which the channel becomes filled with retransmissions and the number of successful packets falls to zero. These topics have only recently been quantified [Metcalfe, 1973a; Lam, 1974] and remain subjects of current investigation.

One approach is to use different constant retransmission intervals at each node, with the intervals equal to integer multiples of the maximum packet transmission time to avoid subsequent conflicts. This results in a priority structure, since nodes assigned the longer intervals will experience a correspondingly longer average delay. As the number of nodes becomes large, however, unacceptably large delays result for the majority of users.

A strategy more appropriate for large user populations is to randomize the retransmission intervals used at each node (note that a priority structure can still be introduced if desired by using larger mean values for lower priority users-in the remaining discussion, equal priorities will be assumed). According to recent results by Lam [1974], the resulting channel behavior appears to be relatively insensitive to the exact nature of the randomization, at least when comparing the use of uniform and geometric distributions. In any event, the cost of implementing a particular distribution at each node is an important design consideration. Based on initial estimates of the expected ALOHANET characteristics, a choice was made to use a uniform distribution. This allowed a relatively simple implementation in both hardware and software user nodes.

A simple technique was used in the original system nodes to achieve short delays when the channel is lightly loaded, while preventing channel saturation from occurring due to peak-hour loading or statistical traffic fluctuations; small retransmission intervals are used (relative to the intervals between new packets), but only for a maximum of three successive retransmission attempts. If the third attempt is unsuccessful, the user is notified of a failure and must manually reinitiate the retransmissions. This in effect introduces a long interval between every three retransmissions, allowing time for retransmissions from other users to succeed. Based on a maximum packet transmission time of 70 milliseconds, the intervals are selected from a range of 0.2 to 1.5 seconds, giving a mean of about 0.7 seconds (ten maximum packet times) per retransmission. The lower bound is chosen to allow sufficient time to receive an ACK from the MENEHUNE if the packet was sent successfully, avoiding unnecessary retransmissions. (This time is based on a direct user-MENEHUNE path; if repeaters form a part of the radio path, the lower limit must be increased accordingly.)

The newer programmable PCU's in the system offer the capability of a more flexible strategy, for example allowing the interval used after each third retransmission to be automatically inserted. The use of different strategies, such as continuously increasing the time range used for selection of successive retransmissions, is also easily implemented by program; these and other strategies are currently under investigation.

Broadcast Channel Queueing

The MENEHUNE acts as a concentrator for the broadcast (F2) channel, queueing waiting traffic when necessary for sequential transmission to user nodes. Four complicating factors exist, however: a need for priority queueing, fair allocation of the channel, the turnaround delay required by half duplex nodes, and the presence of repeaters.

Priority Queues. It is important that the F2 channel data traffic not prevent the prompt return of an ACK to a user node, since this could lead to unnecessary user retransmissions and possible

previous | contents | next