Blocking, reordering, and the throughput of a series of servers (Q793448)

Considered is the throughput of a series of servers, each of which has the same service distribution F with unit mean. It is supposed that the service time of a particular job is the same at each server, and that a job completed at station i cannot advance to station \(i+1\) until station \(i+1\) becomes empty. The same author has previously shown [Adv. Appl. Probab. 14, 633-653 (1982; Zbl 0497.60084)] that the throughput depends on the tail behavior of F, in particular, if service is exponential, the throughput is asymptotically \((\log n)^{-1}.\) The author finds upper and lower bounds on the throughput when the jobs are ordered in blocks of size m, and presented to the system as a block starting with the shortest job. The bounds easily lead to numerical examples, which moreover demonstrate the throughput improvement due to the block ordering. Asymptotic bounds on the throughput are obtained if n,m, and the tail of F obey a certain asymptotic relationship; this makes it possible to design a system to achieve acceptable throughput. Further, it is shown how quickly the block size m must grow in order to maintain asymptotically positive throughput.