Zixuan’s paper on characterizing and modeling non-volatile memory is selected by IEEE Micro Top Picks and will appear in IEEE’s annual “Top Picks from the Computer Architecture Conferences” issue in May/June 2021. Micro Top Picks annually recognizes the best research from architecture conferences. It’s a big honor, so congrats to Zixuan and his co-authors!
The recently released non-volatile memory (NVRAM) products, such as Intel’s Optane DC PM, introduce numerous research opportunities on exploring NVRAM for future computing systems. Recent works discovered NVRAM performance discrepancy compared to conventional DRAM memory. Yet the architecture level details behind this discrepancy remain undiscovered, due to the limited hardware design information revealed by the vendors.
Zixuan’s paper proposes two tools, LENS and VANS, to thoroughly characterize and model the architecture level details of NVRAM. LENS is a low-level non-volatile memory profiler. It runs carefully crafted microbenchmarks that trigger specific hardware behaviors and identifies corresponding hardware properties by analyzing the microbenchmark performance. LENS revealed many hardware components in Intel Optane DIMM, which were not discovered by previous works. VANS is a validated non-volatile memory simulator. It takes observations from LENS and models the NVRAM performance. VANS is accurate, reconfigurable, and can be integrated into full system simulators. You can check these tools on GitHub.