Efficient Memory Page Management for NVDIMM-Based Big Data Processing Environments

Abstract

Due to the recent computing paradigm shift from desktop to cloud environments, memory architectures for big data processing are catching considerable interest. In this paper, we present an efficient page management scheme for large-scale NVDIMM memory. The proposed carefully determines the page size of a large-scale memory such that TLB hit ratio is improved significantly without increasing the page fault ratio, thereby allowing big-data processing efficiently, which was not easy in conventional memory systems. The main contribution of this paper is that it proposes a new paradigm of NVDIMM-based large memory architecture instead of conventional DRAM-based memory system and presents efficient management techniques for that by considering both TLB performances and page fault rates.