A method and apparatus for provide highly efficient traffic routing for a wide range of possible traffic matrices (TM) in an intra-domain network. That routing optimally balances the traffic loads over a range of traffic matrices so as to minimize the deviation for any particular traffic matrix from the optimal routing. Such a routing provides a guaranteed performance ratio against the best possible network routing. The invention utilizes a method of optimally configuring a traffic network based on solving a linear program to obtain the optimal routing, and then configuring the routing on the network accordingly.

A method and apparatus provide improved cache coherency and more effective caching operations without placing an undue burden on network links. A proxy receives a request for a resource and then, depending on information in the proxy cache, generates a resource request for transmission to a resource server. The proxy appends a proxy filter to the request. The resource server maintains one or more volumes of resources based on some predetermined criterion that can be either static or dynamic in nature. Upon receipt of the request and the proxy filter the resource server generates a request response and a piggybacked list of additional resources selected from the volume with which the requested resource is associated.

A method is disclosed for reducing user-perceived latency due to the time required to establish a connection to a server in a network. In accordance with the present invention, an open connection is established to a set of servers, there being some probability the user will contact one of the servers in the near future. This is referred to as preconnecting or prefetching the connection. In the context of a Web client-server network, a list of likely servers can be deduced from links on a current Web page a user is looking at or from a more sophisticated analysis of the user's browsing habits. When the user requests a resource from one of the identified servers, the network connection has already been established, thereby reducing latency and improving service quality, especially for higher bandwidth clients for whom the delay is most noticeable. In contrast to conventional document prefetching, preconnecting does not hog network bandwidth or consume cache space, and hence can be used with much less scrutiny. Moreover, the technique can be implemented in Web browsers without protocol modifications or changes to Web server code.

A method and apparatus provide improved cache coherency and more effective caching operations without placing an undue burden on network links. A proxy receives a request for a resource and then, depending on information in the proxy cache, generates a resource request for transmission to a resource server. The proxy appends a proxy filter to the request. The resource server maintains one or more volumes of resources based on some predetermined criterion that can be either static or dynamic in nature. Upon receipt of the request and the proxy filter the resource server generates a request response and a piggybacked list of additional resources selected from the volume with which the requested resource is associated.

Many pattern recognition tasks, including estimation, classification, and the finding of similar objects, make use of linear models. For example, many text retrieval systems represent queries as linear functions, and retrieve documents whose vector representation has a high dot product with the query. The fundamental operation in such tasks is the computation of the dot product between a query vector and a large database of instance vectors. Often instance vectors which have high dot products with the query are of interest. The invention relates to a random sampling based retrieval system that can identify, for any given query vector, those instance vectors which have large dot products, while avoiding explicit computation of all dot products.