| 摘要 |
<p>Client's (106-1 - 106-N, 107-1 - 107-M) on local area networks (102, 103) making requests to hot sites, which are connected on a wide area network (100) such as the Internet, are redirected through one of a possible plurality of different redirectors (101,103) to one of a possible plurality of caching servers (S1, S2, S3), which each have responsibility for mapping one or more of the hot sites. Each request is probabilistically directed by one of the redirectors to one of the caching servers that map the requested hot site in accordance with weights that are determined for that redirector-hot site pair so as to minimize the average delay that all client requests across the network will encounter in making requests to all the cached hot sites. In order to determine the weights with which each redirector will redirect requests to the hot sites to the caching servers, statistics of access rates to each hot site are dynamically determined by each redirector in the network from the traffic flow and reported to a central management station (CMS) (115). Network delay is similarly measured by each redirector and reported to the CMS, and server delay is computed using a queuing model of each server. Using these parameters as inputs, a non-linear programming optimization problem is solved as a network flow problem in order to determine the weights for each redirector that will minimize the average delay. As the access rate statistics, as well as the network delay and server delay, dynamically change, the CMS, using the network flow algorithm, recalculates the weights and forwards them back to each redirector. In other embodiments, the redirector-logical item pair for which the redirector probabilistically directs client requests may be other than a hot site identity. For example, the logical items can be groups of clients or groups of documents, and the servers to which requests are forwarded can be web servers or caching servers. <IMAGE></p> |
