Computer Science and Engineering, Department of


Date of this Version



University of Nebraska–Lincoln, Computer Science and Engineering
Technical Report TR-UNL-CSE-2008-0008
Issued August 31, 2008


Currently, most generational collectors are tuned to either deliver peak performance when the heap is plentiful, but yield unacceptable performance when the heap is tight or maintain good degradation behavior when the heap is tight, but deliver sub-optimal performance when the heap is plentiful. In this paper, we present NMFLUX (continuously varying the Nursery/Mature ratio), a framework that switches between using a fixed-nursery generational collector and a variable-nursery collector to achieve the best of both worlds; i.e. our framework delivers optimal performance under normal workload, and graceful performance degradation under heavy workload. We use this framework to create two generational garbage collectors and evaluate their performances in both desktop and server settings. The experimental results show that our proposed collectors can significantly improve the throughput degradation behavior of large servers while maintaining similar peak performance to the optimally configured fixed-ratio collector.