Transformer: An OS-Supported Reconfigurable Hybrid Memory Architecture

Non-volatile memories (NVMs) have aroused vast interest in hybrid memory systems due to their promising features of byte-addressability, high storage density, low cost per byte, and near-zero standby energy consumption. However, since NVMs limited write endurance, latency, consumption, it is still challenging directly replace traditional dynamic random access (DRAM) with NVMs. Many studies propose utilize NVM DRAM a system, explore sophisticated management schemes alleviate the impact slow on performance applications. A few architected cache/memory hierarchy. overhead cache metadata (i.e., tags) rather expensive this hierarchical architecture. Some other single (flat) address space form parallel hot page monitoring migration are critical for applications In paper, we Transformer, an OS-supported reconfigurable architecture efficiently use without redesigning hardware To identify frequently accessed (hot) pages migration, count number accesses OSes by sampling bit periodically. We further migrate identified from improve system. More importantly, Transformer can simulate while physically managed flat space, dynamically shift logical between architectures according applications’ patterns. Experimental results show that application 62% average (up 2.7×) compared NVM-only also up 79% 42% (21% 24% average) architectures, respectively.