Capacity management in decentralized networks.

1. Introduction

Bottleneck analysis is a useful tool in capacity planning for centrally controlled network systems, such as job shop or data communication networks. However, under a decentralized network where individual users are allowed to select their own routes, (1) straightforward application of bottleneck analysis does not necessarily yield an optimal performance. In fact, it may even hurt the system performance--an aspect of Braess's paradox. Braess's paradox refers to the phenomenon where adding a new route or capacity results in performance deterioration due to incentive misalignment. Kelly and Cohen (1990) show that a similar phenomenon occurs in ...