Partitioned Persist Ordering

Persistent Memory (PM) technologies enable program recovery to a consistent state in case of failure. To ensure this crash-consistent behavior, programs need to enforce persist ordering by employing mechanisms, such as logging and checkpointing, which introduce additional data movement.The emerging near-data processing (NDP) architectures can effectively reduce this data movement overhead by partitioning the persistent programs and executing the crash consistency mechanisms in the NDP-enabled PM. However, a significant challenge lies in maintaining the persist ordering when execution has been partitioned between the host CPU and NDP-enabled PM. In this work, we first propose Partitioned Persist Ordering (PPO) that ensures a correct persist ordering between CPU and NDP devices, as well as among multiple NDP devices. PPO guarantees high efficiency by reducing unnecessary synchronization among CPU and NDP devices. Based on PPO, we prototype an NDP system, NearPM, on an FPGA platform. NearPM executes data-intensive operations in crash consistency mechanisms with correct ordering guarantees while the rest of the program runs on the CPU. We evaluate nine PM workloads, where each workload supports three crash consistency mechanisms - logging, checkpointing, and shadow paging. Overall, NearPM achieves 4.3-9.8X speedup in the NDP-offloaded operations and 1.22-1.35X speedup in end-to-end execution.