Delay-Free Concurrency on Faulty Persistent Memory

Abstract

Non-volatile memory (NVM) promises persistent main memory that remains correct despite loss of power. This has sparked a line of research into algorithms that can recover from a system crash. Since caches are expected to remain volatile, concurrent data structures and algorithms must be redesigned to guarantee that they are left in a consistent state after a system crash, and that the execution can be continued upon recovery. However, the prospect of redesigning every concurrent data structure or algorithm before it can be used in NVM architectures is daunting.In this paper, we present a construction that takes any concurrent program with reads, writes and CASs to shared memory and makes it persistent, i.e., can be continued after one or more processes fault and have to restart. Importantly the converted algorithm has constant computational delay (preserves instruction counts on each process within a constant factor), as well as constant recovery delay (a process can recover from a fault in a constant number of instructions). We show this first for a simple transformation, and then present optimizations to make it more practical, allowing for a tradeoff for better constant factors in computational delay, for sometimes increased recovery delay. We also provide an optimized transformation that works for any normalized lock-free data structure, thus allowing more efficient constructions for a large class of concurrent algorithms.Finally, we experimentally evaluate transformations by applying them to a queue. We compare the performance of our transformations to that of a persistent transactional memory framework, Romulus, and to a hand-tuned persistent queue. We show that our transformations perform favorably when compared to Romulus. Furthermore, while the hand-tuned version sometimes outperforms our transformations, the difference is not an unreasonable price to pay for the generality and ease of use that we provide.

Brief Announcement: Detectable Sequential Specifications for Recoverable Shared Objects

“…In this paper, we advance the state of the art by presenting a novel formal notion of detectability called the Detectable Sequential Specification (DSS), which is compatible with several existing linearizability-like correctness conditions for recoverable objects [1,6,13], and requires minimal system support. The DSS generalizes and extends the alternative interpretation of NRL proposed by Ben-David, Blelloch, Friedman and Wei [5] in the context of detectable Compare-And-Swap. We further introduce the idea that detectability should be declarative, meaning that the application may request either detectable or non-detectable operations according to its needs.…”

“…As a result, a recovering process suffers from a mild case of amnesia that hinders forensic analysis-it cannot in general determine exactly what step it was about to perform at the point of failure, or what value was returned by its last operation on a given object. Following the approach of [3,5], we treat the amnesia by prescribing the addition of specialized operations to the object's abstract interface. The core idea is to augment a given type with auxiliary operations by which processes declare their intent to execute a detectable operation, and then optionally resolve the status of this operation.…”

FliT: A Library for Simple and Efficient Persistent Algorithms

“…Researchers have also introduced other correctness criteria for persistence and explored how to support them efficiently [6,17,7,4,3]. These works consider not only the state of shared memory upon recovery from a system crash, but also whether processes can continue their previous execution.…”

Sage

“…Persistence is a key property of NVRAMs due to their nonvolatility. Many new persistent data structures have been designed for NVRAMs [7,12,30,31,74,83]. There has also been research on automatic recovery schemes and transactional memory for NVRAMs [3,33,55,59,97,104,108].…”