Times Higher Throughput Research Articles

DF-BETA: An FPGA-based Memory Locality Aware Decision Forest Accelerator via Bit-Level Early Termination

Decision forests, particularly Gradient Boosted Decision Trees (GBDT), are popular due to their high prediction performance and computational efficiency, making them suitable for embedded systems with circuit size and available energy constraints. In this study, we propose a new lightweight GBDT inference acceleration mechanism through the hardware and algorithm co-design. First, we present LoADPack, a hardware-friendly GBDT algorithm that enhances memory access locality. LoADPack obtains trees where the features and thresholds used across the entire ensemble are regular regardless of a branching direction by unifying some nodes and aligning the memory access patterns. Furthermore, we present DF-BETA, a resource-efficient accelerator for the LoADPack algorithm. DF-BETA utilizes MSB-first bit-serial computation to enable early determination of comparison calculations of 32-bit floating-point numbers, optimizing the operation for determining a branch direction. The hardware complexity and computation termination speed vary with the granularity of bit-serial computation. Therefore, we conduct design space exploration of DF-BETA to identify the optimal configuration. Our findings reveal that using 4-bit-serial comparators minimizes circuit size while achieving the leading throughput. Compared to running unconstrained GBDT on a typical accelerator with 32-bit bit-parallel comparators, our accelerator achieves 1.6 times higher throughput on average while maintaining comparable accuracy.

Secure semi‐automated GDPR compliance service with restrictive fine‐grained access control

AbstractSharing personal data with service providers is a contentious issue that led to the birth of data regulations such as the EU General Data Protection Regulation (GDPR) and similar laws in the US. Complying with these regulations is a must for service providers. For users, this compliance assures them that their data is handled the way the service provider says it will be via their privacy policy. Auditing service providers' compliance is usually carried out by specific authorities when there is a need to do so (e.g., data breach). Nonetheless, these irregular compliance checks could lead to non‐compliant actions being undetected for long periods. Users need an improved way to make sure their data is managed properly, giving them the ability to control and enforce detailed, restricted access to their data, in line with the policies set by the service provider. This work addresses these issues by providing a secure semi‐automated GDPR compliance service for both users and service providers using smart contracts and attribute‐based encryption with accountability. Privacy policies will be automatically checked for compliance before a service commences. Users can then upload their personal data with restrictive access controls extracted from the approved privacy policy. Operations' logs on the personal data during its full lifecycle will be immutably recorded and regularly checked for compliance to ensure the privacy policy is adhered to at all times. Evaluation results, using a real‐world organization policy and example logs, show that the proposed service achieves these goals with low time overhead and high throughput.

Hermes, a low-latency transactional storage for binary data streams from remote devices

In many contexts where data is streamed on a large scale, such as video surveillance systems, there is a dual requirement: secure data storage and continuous access to audio and video content by third parties, such as human operators or specific business logic, even while the media files are still being collected. However, using transactions to ensure data persistence often limits system throughput and latency. This paper presents a solution that enables both high ingestion rates with transactional data persistence and near real-time, low-latency access to the stream during collection. This immediate access enables the prompt application of specialized data engineering algorithms during data acquisition. The proposed solution is particularly suitable for binary data sources such as audio and video recordings in surveillance systems, and it can be extended to various big data scenarios via well-defined general interfaces. The scalability of the approach is based on the microservice architecture. Preliminary results obtained with Apache Kafka and MongoDB replica sets show that the proposed solution provides up to 3 times higher throughput and 2.2 times lower latency compared to standard multi-document transactions.

EPO‐R: An efficient garbage collection scheme for long‐term transactions

SummaryThis article proposes EPO‐R, which is an efficient garbage collection scheme designed for multi‐version concurrency control (MVCC) protocols. MVCC generates a version for each update operation, and it can exhaust physical memory for dynamically changing environments such as IoT payments. Eager pruning of obsolete versions (EPO) is a novel garbage collection technique for long‐term transactions. We found room for improvement in EPO, which triggers reclamation invocation. EPO is triggered for each write operation, and it wastes CPU resources. The proposed method EPO‐R is triggered on a read operation to address this issue. The result of experiments with 64 CPU cores and workloads with the combination of short‐term and long‐term transactions demonstrated that EPO‐R exhibited 6.3 times higher throughput than that of EPO. The reason for this can be explained by analyzing the number of unreclaimable versions.

Service-aware real-time slicing for virtualized beyond 5G networks

Edge Intelligence is expected to play a vital role in the evolution of 5G networks, empowering them with the capability to make real-time decisions regarding various allocations related to their management and service provisioning to end-users. This shift facilitates the transition from a network-aware approach, where applications are developed to manage network quality fluctuations, to a service-aware network that self-adjusts based on the hosted applications. In this paper, we design and implement a service-aware network managed from the network edge. We utilize and assess various Machine Learning models to classify cellular network traffic flows in the backhaul, aiming to predict their future impact on network load. Leveraging these predictions, the network can proactively and autonomously reallocate slices in the Radio Access Network via programmable APIs, ensuring the demands of the traffic-generating applications are met. The approach integrates innovative MLOps methodologies for distributed and online training, enabling continuous model refinement and adaptation to evolving network dynamics. Our framework was tested in a real-world environment with realistic traffic scenarios, and the results were evaluated in real-time, down to a granularity of 10ms. Our findings indicate that the network can swiftly adjust to traffic, providing users with slices tailored to their application needs. Notably, our experiments show that under the studied settings, the users experienced up to 4 times lower latency (jitter) and nearly 4 times higher throughput when interacting with various applications, compared to the standard non-AI/ML unit. Furthermore, our dynamic scheme significantly optimizes resource allocation, ensuring energy efficiency by avoiding over- and under-provisioning of resources.

Abstract 4411: Enhancing scalability and consistency in clinical multiomics via an optimized fixed cell ATAC-seq method​

Abstract ATAC-seq has an emerging role in decoding mechanisms of gene regulation, offering valuable insights into pathology and treatment response in disease models. However, clinical adoption of ATAC-seq methods has been limited by logistical hurdles, including time-sensitive processing of fresh samples and compromised viability of cryopreserved cells. These constraints, compounded by changes in open chromatin regions (OCRs) following cryopreservation, introduce unintended bias and pose significant obstacles for the translational impact of ATAC-seq experiments. ​ Here, we introduce an optimized fixed-cell ATAC-seq approach to overcome these limitations and unlock new sample types for ATAC-seq analyses. Our solution improves and simplifies the workflow from sample collection to clinical deliverable. This method enables ATAC-seq investigation of a diverse range of samples and facilitates the execution of complex experimental designs, including time course studies and high throughput screening.​To demonstrate the effectiveness of this method, we compared our optimized fixed-cell method with traditional ATAC-seq preparations in both fresh and cryopreserved GM12878 cells in parallel. Human GM12878 cell line was obtained from Coriell Institute for Medical Research5. Remarkably, we observed consistent genome-wide patterns of OCR enrichment at key regulatory elements across the three sample preparation methods. We observed consistent OCR enrichment across the promoter region of known highly-expressed B-cell genes including CD48 and LCP1, underscoring this assay’s ability to detect chromatin changes at key genes in human disease models. ​ To investigate the potential for multiomic analysis using this method, we prepared RNA-seq libraries from fixed-cell samples in parallel to ATAC-seq. We observed significantly elevated gene expression related to B-cell function and B-cell diseases, demonstrating our method’s compatibility with RNA-seq data collection and integration. This optimized fixed-cell ATAC-seq approach offers enhanced scalability and consistency over conventional methods and presents new opportunities for the multiomic analysis of chromatin and transcriptional activity genome-wide from a single sample in both clinical and research settings. Citation Format: Dafne Alves Winders, Riley Graham, Xiangying Mao, Ilaria De Vito, Andrea O'Hara, Laure Turner, Haythem Latif. Enhancing scalability and consistency in clinical multiomics via an optimized fixed cell ATAC-seq method​ [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4411.

Utility of intercalator displacement assays for screening of ligands for i-motif DNA structures.

Cytosine rich sequences can form intercalated, i-motif DNA structures stabilized by hemi-protonated cytosine:cytosine base pairing. These sequences are often located in regulatory regions of genes such as promoters. Ligands targeting i-motif structures may provide potential leads for treatments for genetic disease. The focus on ligands interacting with i-motif DNA has been increasing in recent years. Here, we describe the fluorescent intercalator displacement (FID) assay using thiazole orange binding i-motif DNA and assess the binding affinity of a ligand to the i-motif DNA by displacing thiazole orange. This provides a time and cost-effective high throughput screening of ligands against secondary DNA structures for hit identification.

BurstSketch: Finding Bursts in Data Streams

<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Burst</i> is a common pattern in data streams which is characterized by a sudden increase in terms of arrival rate followed by a sudden decrease. Burst detection has attracted extensive attention from the research community. To detect bursts accurately in real time, we propose a novel sketch, namely BurstSketch, which consists of two stages. Stage 1 uses the technique Running Track to select potential burst items efficiently. Stage 2 monitors the potential burst items and captures the key features of burst pattern by a technique called Snapshotting. We further propose an optimization, namely Dynamic Buckets, which can improve the accuracy of BurstSketch. We provide theoretical error bounds for Stage 1, Stage 2 and the optimized version. Experimental results show that, compared with the strawman solution, Burstsketch achieves 2.00 to 11.63 times higher F1 score, and 1.56 times higher throughput. We also integrate BurstSketch into Apache Flink, and show that using BurstSketch can be faster than simply using the built-in APIs provided by Apache Flink.

Scalable atomic broadcast: A leaderless hierarchical algorithm

Atomic Broadcast is an essential broadcast primitive as it ensures the consistency of distributed replicas. However, it is notoriously non-scalable. In this work, we introduce the Leaderless Hierarchical Atomic Broadcast (LHABcast) algorithm, which has two properties to improve scalability. First, it is a fully decentralized algorithm that does not rely on a sequencer/leader, which is often a significant bottleneck. Processes running LHABcast send messages with local sequence numbers and order messages received from other processes using timestamps inspired on Lamport's logical clocks. A process that receives the required set of timestamps can make a decision about the overall sequence of message delivery. Second, the algorithm is hierarchical: processes are organized on a vCube logical overlay network, which has several logarithmic properties and allows the construction of autonomous spanning trees. vCube also works as a failure detector, assuming crash faults and an asynchronous system model. In this paper, LHABcast is described, specified, and proven to be correct. Both simulation and experimental results are presented. A comparison with an all-to-all strategy shows that the number of messages sent by LHABcast is significantly lower in both fault-free and faulty scenarios. An implementation of LHABcast in Akka.io achieved up to 3.9 times higher throughput in fault-free scenarios than an implementation of the Raft-based Apache Ratis.

Performance analysis of distributed coordination function with early collision detection in In-band Full-duplex wireless networks

In-band Full-duplex (IB-FD) wireless has the potential to double the spectral efficiency offering its innate capability to resolve conventional problems emerge owing to the imperfect medium access control (MAC) in In-band half-duplex (IB-HD) wireless networks. IEEE 802.11 distributed coordination function (DCF), a binary slotted exponential back-off with carrier sense multiple access scheme, has been used extensively in literature as the fundamental contention resolution technique to devise MAC protocols for IB-FD wireless networks. Unlike IB-HD, an IB-FD network can contain two simultaneously active transmission links within a single collision domain. This phenomenon invalidates the applicability of existing IB-HD models of the DCF to compute its performance (i.e., saturation throughput) accurately in IB-FD wireless networks. We have proposed two MAC protocols for two types of network scenarios taking early collision detection capability of an IB-FD wireless transceiver into consideration. This paper also presents a distinct analytical model to quantify the performance of the DCF in the context of IB-FD wireless assuming finite number of stations and idle channel conditions. We have applied the concept of primary–secondary transmission links and, thoroughly derived the probability that a station transmits either as primary or secondary in a randomly chosen time-slot taking secondary transmitter’s back-off counter into account. We have also investigated the effect of inter-station interference free IB-FD link on the performance of the DCF. Our protocol, when applied to network without hidden stations, has obtained the highest throughput gain of 1.59, 1.61, 1.68 while the DCF is configured with CWmin of 16, 32, 64, respectively than IEEE 802.11 basic access scheme. We have observed increased throughput gain with hidden stations, thanks to inter-station interference free transmission region, and achieved 2.12, 2.14, 2.16 times higher throughput, respectively than that of IEEE rts-cts access scheme.

MorphStream: Adaptive Scheduling for Scalable Transactional Stream Processing on Multicores

Transactional stream processing engines (TSPEs) differ significantly in their designs, but all rely on non- adaptive scheduling strategies for processing concurrent state transactions. Subsequently, none exploit multicore parallelism to its full potential due to complex workload dependencies. This paper introduces MorphStream, which adopts a novel approach by decomposing scheduling strategies into three dimensions and then strives to make the right decision along each dimension, based on analyzing the decision trade-offs under varying workload characteristics. Compared to the state-of-the-art, MorphStream achieves up to 3.4 times higher throughput and 69.1% lower processing latency for handling real-world use cases with complex and dynamically changing workload dependencies.

The Near Infrared Imager and Slitless Spectrograph for JWST. V. Kernel Phase Imaging and Data Analysis

Kernel phase imaging (KPI) enables the direct detection of substellar companions and circumstellar dust close to and below the classical (Rayleigh) diffraction limit. The high-Strehl full pupil images provided by the James Webb Space Telescope (JWST) are ideal for application of the KPI technique. We present a kernel phase analysis of JWST NIRISS full pupil images taken during the instrument commissioning and compare the performance to closely related NIRISS aperture masking interferometry (AMI) observations. For this purpose, we develop and make publicly available the custom Kpi3Pipeline data reduction pipeline enabling the extraction of kernel phase observables from JWST images. The extracted observables are saved into a new and versatile kernel phase FITS file data exchange format. Furthermore, we present our new and publicly available fouriever toolkit which can be used to search for companions and derive detection limits from KPI, AMI, and long-baseline interferometry observations while accounting for correlated uncertainties in the model fitting process. Among the four KPI targets that were observed during NIRISS instrument commissioning, we discover a low-contrast (∼1:5) close-in (∼1 λ/D) companion candidate around CPD-66 562 and a new high-contrast (∼1:170) detection separated by ∼1.5 λ/D from 2MASS J062802.01-663738.0. The 5σ companion detection limits around the other two targets reach ∼6.5 mag at ∼200 mas and ∼7 mag at ∼400 mas. Comparing these limits to those obtained from the NIRISS AMI commissioning observations, we find that KPI and AMI perform similar in the same amount of observing time. Due to its 5.6 times higher throughput if compared to AMI, KPI is beneficial for observing faint targets and superior to AMI at separations ≳325 mas. At very small separations (≲100 mas) and between ∼250 and 325 mas, AMI slightly outperforms KPI which suffers from increased photon noise from the core and the first Airy ring of the point-spread function.

Performance Evaluation of Distributed Database Strategies Using Docker as a Service for Industrial IoT Data: Application to Industry 4.0

Databases are an integral part of almost every application nowadays. For example, applications using Internet of Things (IoT) sensory data, such as in Industry 4.0, are a classic example of an organized storage system. Due to its enormous size, it may be stored in the cloud. This paper presents the authors’ proposition for cloudcentric sensory measurements and measurements acquisition. Then, it focuses on evaluating industrial cloud storage engines for sensory functions, experimenting with three open-source types of distributed Database Management Systems (DBMS); MongoDB and PostgreSQL, with two forms of PostgreSQL schemes (Javascript Object Notation (JSON)-based and relational), against their respective horizontal scaling strategies. Several experimental cases have been performed to measure database queries’ response time, achieved throughput, and corresponding failures. Three distinct scenarios have been thoroughly tested, the most common but widely used: (i) data insertions, (ii) select/find queries, and (iii) queries related to aggregate correlation functions. The experimental results concluded that PostgreSQL with JSON achieves a 5–57% better response than MongoDB for the insert queries (cases of native, two, and four shards implementations), while, on the contrary, MongoDB achieved 56–91% higher throughput than PostgreSQL for the same set up. Furthermore, for the data insertion experimental cases of six and eight shards, MongoDB performed 13–20% more than Postgres in response time, achieving × 2 times higher throughput. Relational PostgreSQL was × 2 times faster than MongoDB in its standalone implementation for selection queries. At the same time, MongoDB achieved 19–31% faster responses and 44–63% higher throughput than PostgreSQL in the four tested sharding subcases (two, four, six, eight shards), accordingly. Finally, the relational PostgreSQL outperformed MongoDB and PostgreSQL JSON significantly in all correlation function experiments, with performance improvements from MongoDB, closing the gap with PostgreSQL towards minimizing response time to 26% and 3% for six and eight shards, respectively, and achieving significant gains towards average achieved throughput.

Reducing Transaction Processing Latency in Hardware Transactional Memory-based Database with Non-volatile Memory

PDF HTML XML Export Cite reminder Reducing Transaction Processing Latency in Hardware Transactional Memory-based Database with Non-volatile Memory DOI: 10.21655/ijsi.1673-7288.00274 Author: Affiliation: Clc Number: Fund Project: National Key Research and Development Program of China (2020YFB2104100); National Science Fund for Distinguished Young Scholars of China (61925206) Article | Figures | Metrics | Reference | Related | Cited by | Materials | Comments Abstract:The emergency of Hardware Transactional Memory (HTM) has greatly boosted the transaction processing performance in in-memory databases. However, the group commit protocol, aiming at reducing the impact from slow storage devices, leads to high transaction commit latency. Non-Volatile Memory (NVM) opens opportunities for reducing transaction commit latency. However, HTM cannot cooperate with NVM together: flushing data to NVM will always cause HTM to abort. In this paper, we propose a technique called parity version to decouple the process of HTM execution and NVM write. Thus, the transactions can correctly and efficiently use NVM to reduce their commit latency with HTM. We have integrated this technique into DBX, a state-of-the-art HTM-based database, and propose DBXN: a low-latency and high-throughput in-memory transaction processing system. Evaluations using typical OLTP workloads including TPC-C show that it has 99% lower latency and 2.1 times higher throughput than DBX. Reference Related Cited by

An FPGA-Based LOCO-ANS Implementation for Lossless and Near-Lossless Image Compression Using High-Level Synthesis

In this work, we present and evaluate a hardware architecture for the LOCO-ANS (Low Complexity Lossless Compression with Asymmetric Numeral Systems) lossless and near-lossless image compressor, which is based on JPEG-LS standard. The design is implemented in two FPGA generations, evaluating its performance for different codec configurations. The tests show that the design is capable of up to 40.5 MPixels/s and 124 MPixels/s per lane for Zynq 7020 and UltraScale+ FPGAs, respectively. Compared to the single thread LOCO-ANS software implementation running in a 1.2 GHz Raspberry Pi 3B, each hardware lane achieves 6.5 times higher throughput, even when implemented in an older and cost-optimized chip like the Zynq 7020. Results are also presented for a lossless only version, which achieves a lower footprint and approximately 50% higher performance than the version that supports both lossless and near-lossless. Interestingly, these great results were obtained applying High-Level Synthesis, describing the coder with C++ code, which tends to establish a trade-off between design time and quality of results. These results show that the algorithm is very suitable for hardware implementation. Moreover, the implemented system is faster and achieves higher compression than the best previously available near-lossless JPEG-LS hardware implementation.

HTAP With Reactive Streaming ETL

In database management systems (DBMSs), query workloads can be classified as online transactional processing (OLTP) or online analytical processing (OLAP). These often run within separate DBMSs. In hybrid transactional and analytical processing (HTAP), both workloads may execute within the same DBMS. This article shows that it is possible to run separate OLTP and OLAP DBMSs, and still support timely business decisions from analytical queries running off fresh transactional data. Several setups to manage OLTP and OLAP workloads are analysed. Then, benchmarks on two industry standard DBMSs empirically show that, under an OLTP workload, a row-store DBMS sustains a 1000 times higher throughput than a columnar DBMS, whilst OLAP queries are more than 4 times faster on a columnar DBMS. Finally, a reactive streaming ETL pipeline is implemented which connects these two DBMSs. Separate benchmarks show that OLTP events can be streamed to an OLAP database within a few seconds.

FlowFight: High performance–low memory top-[formula omitted] spreader detection

A recurring task in security monitoring/anomaly detection applications consists in finding the so-called top “spreaders” (“scanners”), for instance hosts which connect to a large number of distinct destinations or hit different ports. Estimating the top k scanners, and their cardinality, using the least amount of memory meanwhile running at multi-Gbps speed, is a non trivial task, as it requires to “remember” the destinations or ports already contacted in the past by each specific host. This paper proposes and assesses an innovative design, called FlowFight. As the name implies, our approach revolves on the idea of deploying a relatively small number of per-flow HyperLogLog approximate counters — only slightly superior to the target k — and involve the potentially huge number of concurrent flows in a sort of dynamic randomized “competition” for entering such set. The algorithm has been tested and integrated in a full-fledged software router such as Vector Packet Processor. Using either synthetic or real traffic traces, we show that FlowFight is able to estimate the top-k cardinality flows with an accuracy of more than 95%, while retaining a processing throughput of around 8 Mpps on a single core. We further show that FlowFight achieves same accuracy of the state of the art competitor SpreadSketch using 10x times less memory with 1.2x times higher throughput.

Machine-Learned Recognition of Network Traffic for Optimization through Protocol Selection

We introduce optimization through protocol selection (OPS) as a technique to improve bulk-data transfer on shared wide-area networks (WANs). Instead of just fine-tuning the parameters of a network protocol, our empirical results show that the selection of the protocol itself can result in up to four times higher throughput in some key cases. However, OPS for the foreground traffic (e.g., TCP CUBIC, TCP BBR, UDT) depends on knowledge about the network protocols used by the background traffic (i.e., other users). Therefore, we build and empirically evaluate several machine-learned (ML) classifiers, trained on local round-trip time (RTT) time-series data gathered using active probing, to recognize the mix of network protocols in the background with an accuracy of up to 0.96.

Random Channel Access Protocols for SIC Enabled Energy Harvesting IoTs Networks

This article considers channel access in a radio frequency (RF) powered Internet of Things network that consists of a successive interference cancellation (SIC) capable hybrid access point (HAP) and RF-energy harvesting devices. Unlike previous works, in addition to considering random channel access, both the HAP and these devices have imperfect channel state information. We present two distributed Q-learning based channel access strategies to help sensor devices determine their transmission power. Specifically, these two channel strategies exploit conventional and stateless Q-learning. To investigate the influence of centralized and distributed learning on throughput, we further employ the aforementioned channel strategies in the paradigms of independent learner (IL) and joint action learner. The simulation results show that IL outperforms all other strategies. When there is no channel variation, the IL strategy has up to three times higher throughput than Aloha with SIC. The IL strategy can also achieve 100% higher throughput than Aloha with SIC when the frame size is one.

Optimizing TensorFlow Performance by Reconstructing the Convolution Routine

Using deep learning, we can currently build computational models composed of multiple processing layers to learn representations of data. Convolutional neural networks (CNNs) have been widely adopted to achieve significant performance in image recognition and classification. TensorFlow, an open-source deep learning framework from Google, uses profiling to select one convolution algorithm, from among several available, as the core of a CNN to deliver the best performance in terms of execution time and memory usage. However, the overhead from profiling is considerably significant, because TensorFlow executes and profiles all the available algorithms for the best selection whenever an application is launched. We observe that memory usage overshoots during profiling, which limits data parallelism, and thus, fails to deliver maximum performance. In this paper, we present a novel profiling method to reduce overhead by storing the profile result from the first run and reusing it from the second run on. Using Inception-V3, we achieved up to 1.12 times and 1.11 times higher throughput, compared to the vanilla TensorFlow and TensorFlow with XLA JIT compilation, respectively, without losing accuracy.