Performance Evaluation of Intel Optane Memory for Managed Workloads

Intel Optane memory offers non-volatility, byte addressability, and high capacity. It suits managed workloads that prefer large main heaps. We investigate as the for (Java) workloads, focusing on performance scalability. As workload (core count) increases, we note Optane’s relative to DRAM. A few incur a slight slowdown memory, which helps conserve limited DRAM Unfortunately, other scale poorly beyond core counts. This article investigates scaling bottlenecks Java analyzing application, runtime, microarchitectural interactions. Poorly allocate objects rapidly access in frequently. These characteristics slow down mutator substantially garbage collection (GC). At microarchitecture level, load, store, instruction miss penalties rise. To regain performance, partition heaps across hybrid scales considerably better than alone. exploit state-of-the-art GC approaches existing needlessly waste capacity because they ignore runtime behavior. also introduces impact-guided allocation (PIMA) memories. PIMA maximizes utilization, allocating only if it improves performance. estimates impact of either type by sampling. target at graph analytics offering novel estimation method detailed evaluation. identifies phases benefit from with (94.33%) accuracy, incurring 2% sampling overhead. operates stand-alone or combines prior offer new versus trade-offs. work opens up real-life role workloads.