An OS-level Data Distribution Method in DRAM-PCM Hybrid Memory

Abstract

Hybrid memory composed of DRAM and PCM has gained substantial research recently. Compared to each other, DRAM has lower read/write latency and higher endurance, and Phase Change Memory (PCM) has higher density and consumes less energy. Hybrid memory has been proposed to exploit the benefits of both these technologies, while at the same time mitigating their disadvantages. The data distribution methods of state of art approaches were managed by either hardware or compiler, which had some shortcomings. The disadvantage of hardware based approaches is that it need large storage, and the required data swapping degrades overall performance, which is not suitable for certain program which has poor locality. While the compiler based technique requires dynamic program analysis, thus increasing run time overhead and also requires programmer’s help, thus making it a cumbersome approach. We present an OS-level Data Distribution (OSDD) method, in which data sections that have respective read/write features in virtual address space were assigned to different memory medium by memory management module of operating system. Since our approach needs no input from programmer, thus making it transparent. The OSDD based hybrid memory put appropriate data to corresponding memory medium at system level in page granularity and gained better performance and energy saving than former methods, with less overhead. The experiment showed that on average our method get 52 % energy saving at 6 % performance overhead than uniform DRAM memory.