Montage: A General System for Buffered Durably Linearizable Data Structures

Abstract

The recent emergence of fast, dense, nonvolatile main memory suggests that certain long-lived data might remain in its natural pointer-rich format across program runs and hardware reboots. Operations on such data must be instrumented with explicit write-back and fence instructions to ensure consistency in the wake of a crash. Techniques to minimize the cost of this instrumentation are an active topic of research.We present what we believe to be the first general-purpose approach to building buffered durably linearizable persistent data structures, and a system, Montage, to support that approach. Montage is built on top of the Ralloc nonblocking persistent allocator. It employs a slow-ticking epoch clock, and ensures that no operation appears to span an epoch boundary.It also arranges to persist only that data minimally required to reconstruct the structure after a crash. If a crash occurs in epoch e, all work performed in epochs e and e − 1 is lost, but work from prior epochs is preserved.We describe the implementation of Montage, argue its correctness, and report unprecedented throughput for persistent queues, sets/mappings, and general graphs.

Durable Queues: The Second Amendment

“…While the recently-launched Intel Ice Lake processors with Optane persistent memory 200 series may provide flush instructions that do not invalidate the flushed cache lines, existing NVRAM architectures with Cascade Lake processors do not seem to support such instructions. Our impression is corroborated in the findings of [5,15,17,20,28,49,51]. Existing (costly) architectures will probably remain in use for years to come and one needs to use algorithmic modifications to obtain improved performance on such machines.…”

“…This is also noted by others [e.g. 5,15,17,20,28,49,51]. The recently-launched Third Generation Intel Xeon Scalable Ice Lake processors with Optane persistent memory 200 series may implement CLWB retaining lines in the cache, but NVRAM platforms currently in the market do not seem to support flushes without cache invalidation.…”

Fast Nonblocking Persistence for Concurrent Data Structures

“…Both Dalí and InCLL employ techniques that might be applicable in principle to other data structures. To the best of our knowledge, however, the Montage system of Wen et al [54] is the only existing general-purpose system for buffered durable linearizable structures. It also has the advantage of persisting only semantically essential data.…”

“…Other data may reside in transient memory. The original system API [54] is designed primarily for lock-based data structures, but also includes support for nonblocking operations (with blocking advance of the persistence frontier). The typical operation is bracketed by calls to begin op and end op.…”

“…Inspired by Dalí and the InCLL MassTree, Wen et al [54] introduced the first and, to our knowledge, only general purpose system for buffered durably linearizable data structures. Like the transient-index sets of the previous paragraph, their Montage system also facilitates persisting only semantically essential payloads.…”