EFFICIENTLY PROVIDING MULTIPLE GRADES OF SERVICE WITH PROTECTION AGAINST OVERLOADS IN SHARED RESOURCES

摘要

Techniques for (a) controlling admission of customers to a shared resource, (b) adjusting the capacity of a resource in light of new customer demand, and (c) diverting usage from a failed resource to alternative resources, each use a "blocking probability computer" (BPC) to solve a resource-sharing model that has a product-form steady-state distribution. The techniques allow each customer to obtain an appropriate grade of service and protection against overloads from other customers. Each customer is a source of a series of requests, and is assigned "upper-limit" (UL) and "guaranteed-minimum" (GM) "bounds" on its requests. The upper limit bound puts an upper limit on the number of requests from that customer that can be in service at any time. The guaranteed-minimum bound guarantees thatthere will always be available resource units in the resources to serve a specified number of requests from that customer. The desired blocking probabilities are directly expressed in terms of normalization constants appearing in the product-form steady-state distribution. The BPC computes the normalization constants by firstconstructing the generating function (or z-transform) of the normalizing constant and then numerically inverting the generating function.