tabate writes: Stanford researchers Christina Delimitrou and Christos Kozyrakis have developed a cluster management system called Quasar that significantly boosts datacenter utilization. Current utilization is notoriously low, primarily because users that submit applications to these systems are extremely conservative when it comes to reserving resources, often overprovisioning reservations by an order of magnitude or more to side-step performance unpredictability issues. Quasar takes a different approach which requires the user to specify a performance target for an application (e.g., 99th percentile latency) and translates that into resources using an approach similar to online recommendation systems like Netflix or Amazon's. By projecting a small profiling signal for a new workload against the massive dataset assembled from previously-scheduled applications, Quasar can find similarities between resource preferences of seemingly non-similar workloads. This way resource allocations are sized much more tightly to just meet the performance requirements of incoming workloads and as a result datacenter utilization improves without sacrificing application performance.