Online fundraising with group interaction and information sharing transforms traditional individual private donation information into group-visible information, which further affects the online donation behaviour of individuals receiving donation information from others in the fundraising platform. Based on the type of donation information from others and the relationship strength between individuals and others, this study analyses the changes in the donation intention of individuals receiving beneficial or damaging donation information from others with different levels of relationship strength. The results of multi-situation simulation experiments reveal that, the beneficial donation information of others with weak (strong) relationship strength will arouse individuals’ low (high) group identity, making their online donation amount close to (far exceeding) the platform’s recommended donation benchmark, with strong (weak) donation autonomy. The damaging donation information of others with weak (strong) relationship strength will arouse individuals’ low (high) deontic justice, making their online donation amount less than (far exceeding) the platform’s recommended donation benchmark, with weak (strong) donation autonomy. Research conclusions further enrich the theory of public welfare behaviour from the online context and group level, and provide inspiration for emerging online platforms to optimise the design of fundraising activities and promote individual rational donations.

Deep learning recommendation models (DLRMs) have been used across many business-critical services at Meta and are the single largest AI application in terms of infrastructure demand in its data-centers. In this paper, we present Neo, a software-hardware co-designed system for high-performance distributed training of large-scale DLRMs. Neo employs a novel 4D parallelism strategy that combines table-wise, row-wise, column-wise, and data parallelism for training massive embedding operators in DLRMs. In addition, Neo enables extremely high-performance and memory-efficient embedding computations using a variety of critical systems optimizations, including hybrid kernel fusion, software-managed caching, and quality-preserving compression. Finally, Neo is paired with ZionEX, a new hardware platform co-designed with Neo's 4D parallelism for optimizing communications for large-scale DLRM training. Our evaluation on 128 GPUs using 16 ZionEX nodes shows that Neo outperforms existing systems by up to 40× for training 12-trillion-parameter DLRM models deployed in production.

Despite recognizing the importance of virtual brand community (VBC) and the existence of different types of electronic word-of-mouth (eWOM) behaviors (e.g., opinion giving, opinion passing and opinion seeking), the existing literature remains inadequate regarding the influences of VBC experiences and identification on eWOM. This study, incorporating stimulus–organism–response theory and related research, examines the influences of various VBC experiences (e.g., intellectual, sensory, entertainment, interactive, affective, and relational experiences) on the different types of eWOM behaviors, mediated via two different types of identification (i.e., brand identification and community identification). Data are collected using an online survey and analyzed using structural equation modeling. Results show that VBC experiences (with the exception of interactive experience) have a significant impact on brand identification and community identification. Moreover, community identification and brand identification distinctively affect the different types of eWOM behaviors. Theoretical and practical implications, limitations, and future research directions are discussed.

At Facebook, machine learning provides a wide range of capabilities that drive many aspects of user experience including ranking posts, content understanding, object detection and tracking for augmented and virtual reality, speech and text translations. While machine learning models are currently trained on customized data-center infrastructure, Facebook is working to bring machine learning inference to the edge. By doing so, user experience is improved with reduced latency (inference time) and becomes less dependent on network connectivity. Furthermore, this also enables many more applications of deep learning with important features only made available at the edge. This paper takes a data-driven approach to present the opportunities and design challenges faced by Facebook in order to enable machine learning inference locally on smart phones and other edge platforms.

Data ingestion is an essential part of companies and organizations that collect and analyze large volumes of data. This paper describes Gobblin, a generic data ingestion framework for Hadoop and one of LinkedIn's latest open source products. At LinkedIn we need to ingest data from various sources such as relational stores, NoSQL stores, streaming systems, REST endpoints, filesystems, etc. into our Hadoop clusters. Maintaining independent pipelines for each source can lead to various operational problems. Gobblin aims to solve this issue by providing a centralized data ingestion framework that makes it easy to support ingesting data from a variety of sources. Gobblin distinguishes itself from similar frameworks by focusing on three core principles: generality, extensibility, and operability. Gobblin supports a mixture of data sources out-of-the-box and can be easily extended for more. This enables an organization to use a single framework to handle different data ingestion needs, making it easy and inexpensive to operate. Moreover, with an end-to-end metrics collection and reporting module, Gobblin makes it simple and efficient to identify issues in production.

Espresso is a document-oriented distributed data serving platform that has been built to address LinkedIn's requirements for a scalable, performant, source-of-truth primary store. It provides a hierarchical document model, transactional support for modifications to related documents, real-time secondary indexing, on-the-fly schema evolution and provides a timeline consistent change capture stream. This paper describes the motivation and design principles involved in building Espresso, the data model and capabilities exposed to clients, details of the replication and secondary indexing implementation and presents a set of experimental results that characterize the performance of the system along various dimensions.

Linked In is among the largest social networking sites in the world. As the company has grown, our core data sets and request processing requirements have grown as well. In this paper, we describe a few selected data infrastructure projects at Linked In that have helped us accommodate this increasing scale. Most of those projects build on existing open source projects and are themselves available as open source. The projects covered in this paper include: (1) Voldemort: a scalable and fault tolerant key-value store, (2) Data bus: a framework for delivering database changes to downstream applications, (3) Espresso: a distributed data store that supports flexible schemas and secondary indexing, (4) Kafka: a scalable and efficient messaging system for collecting various user activity events and log data.

This paper presents a power-efficient parallel cache coherence protocol for network-on-chip interconnection fabric on many core chips. With the increasing numbers of processor cores, a directory-based cache coherence protocol is more scalable and expected to be used for the future chip architectures. However, the characterization of directory-based protocol on a NoC platform shows that many cycles are used to perform command communications to ensure the data consistency before a data packet is transmitted or processed, which restricts the performance. In this paper, a parallel cache coherence protocol is designed to address the problem, which decouples the transmission of the data packets and command packets. Specifically, the parallel mechanism also provides an opportunity to reduce power consumption by using power-optimized wires. A formal speedup model has been established to estimate the performance of this approach. Simulation experiments show significant performance improvement and good system scalability: (1) the average latency reductions are 4.38%, 5.35%, 8.53% and 11.25% for 16, 32, 64, and 128-cores, respectively, (2) the numbers of the L2 cache accesses are greatly reduced, (3) the proposed protocol improves the system performance with up to 18.6%, and (4) the average power savings are 5.66%, 8.14%, 10.73%, and 13.23% for 16, 32, 64 and 128-cores, respectively.

This paper presents a fast and cycle-accurate memory subsystem modeling and evaluating framework for Chip Multiprocessors (CMPs), called TSIM (Tsinghua SIMulator), which gives a flexible and extensible approach to evaluating architecture designs, models or algorithms, including the network-on-chip interconnection, cache hardware prefetcher, memory system protocol, replacement policy, etc. TSIM is trace-driven, adopting a dynamic binary instrumentation technique to generate the running trace information of applications on-the-fly. After receiving the trace information, TSIM will reappear the on-chip memory behaviors of applications. By introducing the concept of statistical meta metrics, TSIM separates the analysis stage from the simulation process per se, and this provides a great facilitation for a user to count and sample the performance metrics. Compared to the real cache system, TSIM achieves an accuracy of 90.66% at the average speed of 327 KIPS. Meanwhile, TSIM accelerates the simulating speed by almost 10 times, compared to the traditional cycle-accurate cache simulators. On the other hand, when TSIM is used to characterize the on-chip memory system behaviors of SPEC CPU 2000 benchmarks, experimental results about the on-chip memory behaviors are the same as others.