HP Labs Technical Reports



Click here for full text: PDF

A Simulation Study of Generalised Processor Sharing

Wood, Jamie; O'Connell, Neil


HPL-BRIMS-96-23

Keyword(s): No keywords available.

Abstract: In recent work of the second author, some curious phenomena were observed concerning optimal resource allocation in a resource with two incoming traffic streams with dedicated storage space in a finite buffer, and limited output capacity to be shared between the two streams according to some priority rule. An example is given where there are two independent incoming streams, with identical statistical properties, yet the optimal allocation of resources, from the point of view of minimising the overall frequency of overflow, is to give top priority to either of the two streams and more buffer space to the other. These results were obtained using large deviation theory and as such are only large buffer approximations; also, explicit calculations were only possible for relatively specific traffic types. In this paper we explore, by simulation, the universality of these phenomena. We find that, not only do the observations persist for a wide variety of traffic types, they are also valid for systems with small buffers. There is a 'rotational' effect as the buffer size decreases, leading to optimal policies that give top priority to one stream, as before, but that stream gets an even smaller share of the available buffer space.

Back to Index

[Research] [News] [Tech Reports] [Palo Alto] [Bristol] [Japan] [Israel] [Site Map] [Home] [Hewlett-Packard]