persistentmemory

Persistent memory refers to non-volatile memory technologies that retain data after power is removed. In computing, the term commonly describes byte-addressable memory that can be accessed directly by the processor as primary memory while also maintaining contents across power cycles. This combination enables programming models where data structures reside in memory with durability guarantees, reducing or eliminating the need for conventional storage I/O on many workloads.

Persistent memory technologies include phase-change memory, magnetoresistive RAM, and other non-volatile memories, with Intel Optane DC Persistent Memory being a well-known commercial example. These can be deployed as memory DIMMs in two main modes: a memory mode, which hides persistence from applications, and app-direct mode, which treats the memory as durable, byte-addressable storage that can be accessed with persistence semantics. Performance is typically faster than traditional storage but may fall short of DRAM, and endurance and cost considerations influence deployment.

Software support focuses on providing crash-consistent programming models and durability guarantees. The Persistent Memory Development Kit (PMDK) offers libraries such as libpmem, libpmemobj, and libpmemlog to build persistent data structures and transactions. Operating systems and file systems can expose persistent memory via direct access mappings, enabling data to remain after a crash without conventional filesystem I/O.

Applications include databases, in-memory analytics, graph processing, and fast recovery after crashes. While persistent memory promises substantial speedups and simplified data durability, it introduces challenges related to cost, endurance, data consistency, and programming complexity. Ongoing development aims to broaden adoption and simplify software stacks.