Speaker
Description
New memory technologies are an emerging to provide larger RAM sizes at reasonable cost and energy consumption. In addition to the conventional DRAM, recent memory infrastructures contain byte-addressable persistent memory (PMEM) technology that offers capacities higher than DRAM and better access times than Nand-based technologies such as SSDs.
In such hybrid infrastuctures, users have the choice to either manually manage allocations to different memory spaces or delegate such a task to system components where DRAM is used as a cache for PMEM. Nevertheless, recent research showed that both approaches have limitations. From the one hand, hardware-based mechanisms lack flexibility and are not always efficient, yielding inconsistent performance. From the other hand, software-based approaches pose management overheads and often require expert knowledge and intrusive code changes.
Hence, in order to fully take advantage of different capabilities offered by hybrid memory systems and remove the memory management burden from the user, we introduce the EcoHMEM methodology for automatic object-level placement. The methodology of the EcoHMEM frameowrk is able to transparently optimize applications memory allocations without making any modifications to the source codes and without requiring any modifications to the kernel. Compared to state–of–the–art solutions, our methodology can attain up to 2x runtime improvement in mini-benchmarks and up-to 6% improvement in complex production applications.
