Key-value stores such as LevelDB and RocksDB offer excellent write throughput, but suffer high amplification. The amplification problem is due to the Log-Structured Merge Trees data structure that underlies these key-value stores. To remedy this problem, paper presents a novel inspired by Skip Lists, termed Fragmented (FLSM). FLSM introduces notion of guards organize logs, avoids rewriting in same level. We build PebblesDB, high-performance store, modifying HyperLevelDB use structure....

Social robots have to potential serve as personal, organizational, and public assistants as, for instance, diet coaches, teacher's aides, emergency respondents. The success of these - whether in motivating users adhere a regimen or encouraging them follow evacuation procedures the case fire will rely largely on their ability persuade people. Research range areas from political communication education suggest that nonverbal behaviors human speaker play key role persuasiveness speaker's...

We introduce optimistic crash consistency, a new approach to consistency in journaling file systems. Using an array of novel techniques, we demonstrate how build commit protocol that correctly recovers from crashes and delivers high performance. implement this within Linux ext4 variant which call OptFS. two file-system primitives, osync() dsync(), decouple ordering writes their durability. show through experiments OptFS improves performance for many workloads, sometimes by order magnitude;...

We present Recipe, a principled approach for converting concurrent DRAM indexes into crash-consistent persistent memory (PM). The main insight behind Recipe is that isolation provided by certain class of in-memory can be translated with small changes to crash-consistency when the same index used in PM. set conditions enable identification this indexes, and actions taken convert each persistent. Based on these conversion actions, we modify five different based B+ trees, tries, radix hash...

Modern file systems use ordering points to maintain consistency in the face of system crashes. However, such leads lower performance, higher complexity, and a strong perhaps naive dependence on layers correctly enforce writes. In this paper, we introduce No-Order File System (NoFS), simple, lightweight that employs novel technique called backpointer-based provide crash without writes as they go disk. We utilize formal model prove NoFS provides data event crashes; show through experiments is...

We present the first comprehensive study of application-level crash-consistency protocols built atop modern file systems. find that applications use complex update to persist state, and correctness these is highly dependent on subtle behaviors underlying system, which we term persistence properties. develop a tool named BOB empirically tests properties, it demonstrate properties vary widely among six popular Linux build framework ALICE analyzes application finds crash vulnerabilities, i.e.,...

Training Deep Neural Networks (DNNs) is resource-intensive and time-consuming. While prior research has explored many different ways of reducing DNN training time, the impact input data pipeline , i.e., fetching raw items from storage performing pre-processing in memory, been relatively unexplored. This paper makes following contributions: (1) We present first comprehensive analysis how affects time widely-used computer vision audio (DNNs), that typically involve complex pre-processing....

The General Data Protection Regulation (GDPR) provides new rights and protections to European people concerning their personal data. We analyze GDPR from a systems perspective, translating its legal articles into set of capabilities characteristics that compliant must support. Our analysis reveals the phenomenon metadata explosion, wherein large quantities needs be stored along with data satisfy requirements. also helps us identify workloads supported under GDPR. design implement an...

We present SplitFS, a file system for persistent memory (PM) that reduces software overhead significantly compared to state-of-the-art PM systems. SplitFS presents novel split of responsibilities between user-space library and an existing kernel system. The handles data operations by intercepting POSIX calls, memory-mapping the underlying file, serving read overwrites using processor loads stores. Metadata are handled (ext4 DAX). introduces new primitive termed relink efficiently support...

Modern persistent-memory (PM) file systems perform well in benchmark settings, when the system is freshly created and empty. But after being aged by usage, as will be normal mode practice, their memory-mapped performance degrades significantly. This paper shows that cause inability to use 2MB hugepages map files aged, having 4KB pages instead suffering many extra page faults TLB misses a result.

We present Dinomo, a novel key-value store for disaggregated persistent memory (DPM). Dinomo is the first DPM that simultaneously achieves high common-case performance, scalability, and lightweight online reconfiguration. observe previously proposed stores had architectural limitations prevent them from achieving all three goals simultaneously. uses combination of techniques such as ownership partitioning, adaptive caching, selective replication, lock-free log-free indexing to achieve these...

Blizzard is a high-performance block store that exposes cloud storage to cloud-oblivious POSIX and Win32 applications. connects clients servers using network with full-bisection bandwidth, allowing access any remote disk as fast if it were local. Using novel striping scheme, high parallelism both sequential random workloads; also, by decoupling the durability ordering requirements expressed flush requests, can commit writes out-of-order, providing performance crash consistency applications...

Recent research has shown that applications often incorrectly implement crash consistency. We present the Crash-Consistent File System (ccfs), a file system improves correctness of application-level consistency protocols while maintaining high performance. A key idea in ccfs is abstraction stream . Within stream, updates are committed program order, improving correctness; across streams, there no ordering restrictions, enabling scheduling flexibility and empirically demonstrate running atop...

We present a new approach to testing file-system crash consistency: bounded black-box (B3). B3 tests the file system in manner using workloads of operations. Since space possible is infinite, bounds this based on parameters such as number operations or which include, and exhaustively generates within space. Each workload tested target by simulating power-loss crashes while being executed, checking if recovers correct state after each crash. builds upon insights derived from our study...

Mesenteric ischaemia comprises a broad, heterogeneous group of diseases characterised by inadequate blood supply to the small or large bowel. Acute mesenteric is surgical emergency, with significant associated morbidity and mortality. Because clinical presentation variable often nonspecific, high index radiologic suspicion required for early diagnosis. The severity ranges from transient, localised frank necrosis most common causes acute are embolic thrombotic occlusion superior artery,...

Most modern data processing pipelines run on top of a distributed storage layer, and securing the whole system, layer in particular, against accidental or malicious misuse is crucial to ensuring compliance rules regulations. Enforcing protection privacy rules, however, stands at odds with requirement achieve higher access bandwidths rates large pipelines. In this work we describe our proposal for path forward that reconciles two goals. We call approach "Software-Defined Data Protection"...

Abstract One of the most frequently used indirect techniques for deriving estimates recent fertility from simple questions in censuses and surveys is ‘P/F ratio’ method. Availability detailed birth-history data, as World Fertility Survey, applications P/F procedure a diagnostic tool evaluation quality data have led to simplifications extensions original This analysis illustrates that when complete maternity histories are available, can be simplified made more powerful by (1) calculation...

Rethinking the fundamental abstractions of file system.