Using queueing models and GPSS simulation, the author investigates whether, in an ARPANET-like packet switch network, network throughput is improved when acknowledgement packets are given a higher priority than normal traffic. The study is motivated by statistics gathered from ARPANET in 1982–1983, showing that over 95 percent of the traffic consists of single-packet, individually acknowledged messages. The author also investigates the effects of a shortest-packet-first policy, under which a no preemption case allows short packets in the send queues to get high priority, while a with preemption case allows short incoming packets to interrupt longer outgoing ones.
Results for several selected types of traffic flows indicate that prioritization has marginal or inconclusive impact; far more significant in affecting throughput is the window size (the number of unacknowledged packets allowed). In fact, as the author points out, datagrams may be more appropriate than acknowledged virtual circuits in an environment so rich with single-packet traffic.
The paper is marred by a host of irritating syntax and spelling flaws that should have been detected in editing, including the repeated use of a connecting dash where a separating dash is meant and several instances where principle is used instead of principal.