Data Cache Research Articles

PSAC: Proactive Sequence-Aware Content Caching via Deep Learning at the Network Edge

Compared with traditional ineffective methods, such as acquiring more spectrum and deploying more base stations, edge caching is a highly promising solution for increased data flow needs and has attracted considerable attention. However, owing to the lack of careful consideration of cached data, existing related methods neither reduce network load nor improve the quality of experience. In this study, we propose a proactive sequence-aware content caching strategy (PSAC). Specifically, for general content at the network edge and content with sequential features, PSAC_gen (based on a convolutional neural network) and PSAC_seq (based on an attention mechanism that can automatically capture sequential features), respectively, are proposed to implement proactive caching. Experiments demonstrate that the proposed deep learning content caching method can effectively improve user experience and reduce network load.

OMCPR: Optimal Mobility Aware Cache Data Pre-fetching and Replacement Policy Using Spatial K-Anonymity for LBS

Location-based services are an important category of context-aware computing, which play an important role in providing the continuous, local and spatially confined information systems very efficiently and accurately to their clients. The key component in location-based services is the current location of a mobile user. Thus, to protect mobile users’ location and their other information to an untrusted party with consideration of minimal waiting time, Optimal Mobility Aware Cache data Pre-fetching and Replacement policy (OMCPR) is being proposed here. In this continuous location-based services model, the system introduces a mediator namely Anonymizer by employing the prefetching facility for spatial K-anonymity that resides in between the user and Query Analyser to form a cloaking region using mobile user inputs (data freshness, the contribution rate of cell’s cache and location). It provides high-quality lossless location-based services by the utilization of the frequent pattern mining of mobile users’ trajectories to forecast their next position as per mobility and multiple constraints. A client–server based queueing model is used to simulate the proposed OMCPR platform. It provides higher privacy protection than the current state of the art strategies available, also minimizes the overhead of the LBS server and waiting time of mobile users by the addition of the prefetching facilities to Anonymizer.

Optimizing synchronization mechanism for block-based file systems using persistent memory

Existing block-based file systems employ buffer caching mechanism to improve performance, which may result in data loss in the case of power failure or system crash. To avoid data loss, the file systems provide synchronization operations for applications to synchronously write the dirty data in DRAM cache back to the slow block devices. However, the synchronization operations can severely degrade the performance of the file system since violating the intention of buffer caching mechanism. In this paper, we propose to relieve the overhead of synchronization operations while ensuring data reliability by utilizing a small Persistent Memory. The proposed Persistent Memory assisted Write-back (PMW) mechanism includes a dedicated Copy-on-Write mechanism to guarantee data consistency and a write-back mechanism across PM and the block devices. We implement the proposed PMW in Linux kernel based on Ext4. The experimental results show that PMW can achieve about 2.2× and 1.6× performance improvement over the original Ext4 and AFCM, the state-of-the-art PM-based synchronization mechanism, on the TPCC workload, respectively.

Speech and web-based technology to enhance education for pupils with visual impairment

This paper describes a new web-based system specially adapted to the education of Czech pupils with visual impairment. The system integrates speech and language technologies with a web framework in lower secondary education, especially in mathematics and physics subjects. A new interface utilized the text-to-speech (TTS) synthesis for online automatic reading of educational texts. The interface provides several TTS voices, synthesized data caching, and automatic processing of formulas in mathematics and physics. The system was designed to enable teachers create and manage teaching materials. It also enables the pupils to view and listen to the read forms of these documents online. A school for pupils with visual impairment participated in the development and implementation of the system. After one year of using the system daily, the user experience and evaluation data were collected. The results indicate a positive reception and frequent use of the system as well as a preference over classical educational materials.

Using DODAS as deployment manager for smart caching of CMS data management system

DODAS stands for Dynamic On Demand Analysis Service and is a Platform as a Service toolkit built around several EOSC-hub services designed to instantiate and configure on-demand container-based clusters over public or private Cloud resources. It automates the whole workflow from service provisioning to the configuration and setup of software applications. Therefore, such a solution allows using “any cloud provider”, with almost zero effort. In this paper, we demonstrate how DODAS can be adopted as a deployment manager to set up and manage the compute resources and services required to develop an AI solution for smart data caching. The smart caching layer may reduce the operational cost and increase flexibility with respect to regular centrally managed storage of the current CMS computing model. The cache space should be dynamically populated with the most requested data. In addition, clustering such caching systems will allow to operate them as a Content Delivery System between data providers and end-users. Moreover, a geographically distributed caching layer will be functional also to a data-lake based model, where many satellite computing centers might appear and disappear dynamically. In this context, our strategy is to develop a flexible and automated AI environment for smart management of the content of such clustered cache system. In this contribution, we will describe the identified computational phases required for the AI environment implementation, as well as the related DODAS integration. Therefore we will start with the overview of the architecture for the pre-processing step, based on Spark, which has the role to prepare data for a Machine Learning technique. A focus will be given on the automation implemented through DODAS. Then, we will show how to train an AI-based smart cache and how we implemented a training facility managed through DODAS. Finally, we provide an overview of the inference system, based on the CMS-TensorFlow as a Service and also deployed as a DODAS service.

Automated predictions from polarized matrix elements

The anticipated experimental resolution and data cache of the High Luminosity Large Hadron Collider will enable precision investigations of polarization in multiboson processes. This includes, for the first time, vector boson scattering. To facilitate such studies, we report the automation of polarized matrix element computations in the publicly available Monte Carlo tool suite, MadGraph5_aMC@NLO. This enables scattering and decay simulations involving helicity-polarized asymptotic or intermediate states, preserving both spin-correlation and off-shell effects. As demonstrations of the method, we investigate the leading order production and decay of polarized weak gauge bosons in the process ppto {jjW}_{lambda}^{+}{W}_{lambda prime}^{-} , with helicity eigenstates (λ, λ′) defined in various reference frames. We consider the Standard Model at both mathcal{O} (α4) and mathcal{O} (α2 {alpha}_s^2 ) as well as a benchmark composite Higgs scenario. We report good agreement with polarization studies based on the On-Shell Projection (OSP) technique. Future capabilities are discussed.

Performance Assessment of RISC-V Architecture

This paper presents the performance evaluation of RISC – V architecture based processor using Gem5 simulator. The performance analysis metrics such as bandwidth, latency, throughput, branch prediction, pipeline stages and memory hierarchy of the processor architecture are studied using Gem5 simulator. Different simulation models are carried out to arrive the best reference model for RISC-V architecture design and development. In this reference model cache memory functionality feature is verified with the verification methodology called Universal Verification Methodology (UVM). From simulations it is found that both the program and data cache provides optimum performance in terms of execution time, hit rates, miss rate and miss latencies.

Distributed Data Access in Industrial Edge Networks

Wireless edge networks in smart industrial environments increasingly operate using advanced sensors and autonomous machines interacting with each other and generating huge amounts of data. Those huge amounts of data are bound to make data management (e.g., for processing, storing, computing) a big challenge. Current data management approaches, relying primarily on centralized data storage, might not be able to cope with the scalability and real time requirements of Industry 4.0 environments, while distributed solutions are increasingly being explored. In this paper, we introduce the problem of distributed data access in multi-hop wireless industrial edge deployments, whereby a set of consumer nodes needs to access data stored in a set of data cache nodes, satisfying the industrial data access delay requirements and at the same time maximizing the network lifetime. We prove that the introduced problem is computationally intractable and, after formulating the objective function, we design a two-step algorithm in order to address it. We use an open testbed with real devices for conducting an experimental investigation on the performance of the algorithm. Then, we provide two online improvements, so that the data distribution can dynamically change before the first node in the network runs out of energy. We compare the performance of the methods via simulations for different numbers of network nodes and data consumers, and we show significant lifetime prolongation and increased energy efficiency when employing the method which is using only decentralized low-power wireless communication instead of the method which is using also centralized local area wireless communication.

Position-Based Adaptive Clustering Model (PACM) for Efficient Data Caching in Vehicular Named Data Networks (VNDN)

The seamless data delivery is essential in VANET for application such as autonomous vehicle, intelligent traffic management and for the road safety and emergency applications. The incorporation of named data networking (NDN) with VANET, intended to frame intelligent traffic flow and seamless data delivery. Such integration of vehicular ad hoc networks (VANET) with NDN is termed as vehicular named data networks (VNDN). Because of the continuous node/vehicle mobility, it is a tedious process to build constant and consistent communication between vehicles. With that concern, for enhancing the performance of VNDN and solving the issues such as frequent cluster formation on heavy loaded data transmissions, position-based adaptive clustering model (PACM) is developed. The major intention of PACM is to form clusters based on trajectory. Besides, PACM performs efficient data caching by collecting significant data from vehicles to establish consistent data communication with all nodes in the network. Efficient data caching is done with the elected cluster heads among the framed clusters based on its positions and mobility models. For handling the vehicles at higher mobility speed, mutual data caching process is also designed that makes vehicles to perform on-demand data gathering from cluster heads. Further, the model is simulated and the obtained results are compared with the existing models based on the metrics such as packet delivery ratio, mean delay, cache hit rate and mean hop distance. The comparative analysis shows that the proposed model outperforms the available techniques.

Optimal Energy Consumption for Communication, Computation, Caching, and Quality Guarantee

Energy efficiency is a fundamental requirement of modern data-communication systems, and its importance is reflected in much recent work on performance analysis of system energy consumption. However, most work has only focused on communication and computation costs without accounting for data caching costs. Given the increasing interest in cache networks, this is a serious deficiency. In this paper, we consider the problem of energy consumption in data communication, computation and caching (C3) with a quality-of-information (QoI) guarantee in a communication network. Our goal is to identify the optimal data compression rates and cache placement over the network that minimizes the overall energy consumption in the network. We formulate the problem as a mixed integer nonlinear programming (MINLP) problem with nonconvex functions, which is non-deterministic polynomial-time hard (NP-hard) in general. We propose a variant of the spatial branch-and-bound algorithm (V-SBB) that can provide an $\epsilon$-global optimal solution to the problem. By extensive numerical experiments, we show that the C3 optimization framework improves the energy efficiency by up to 88% compared to any optimization that only considers either communication and caching or communication and computation. Furthermore, the V-SBB technique provides comparatively better solutions than some other MINLP solvers at the cost of additional computation time.

Design of a Digital Readout Module for SCA ASIC Testing

Switched Capacitor Array (SCA) features high-speed sampling and low power consumption. High-speed waveform digitization based on SCA is an important research direction for high-precision time measurement. We are designing the SCA ASIC chips, of which several prototypes have been designed and the improvements are ongoing. In order to test and evaluate the future SCA ASIC chips, it is necessary to design a digital readout module with good compatibility. This paper presents the design of the readout module and the data readout software. The digital readout module integrates the functionality of the control, configuration and data readout of the ASIC under test within one FPGA device. A DDR3 chip is used for data caching, and a USB3.0 interface is designed for data transmission. The readout software is designed based on Python3.7 and contains the functions such as data acquisition and waveform drawing. This module was successfully used in the initial testing of the second version of the SCA prototype ASIC, and the test results indicate that the readout module works as expected.

Architecture of a distributed storage that combines file system, memory and computation in a single layer

Storage and memory systems for modern data analytics are heavily layered, managing shared persistent data, cached data, and non-shared execution data in separate systems such as a distributed file system like HDFS, an in-memory file system like Alluxio, and a computation framework like Spark. Such layering introduces significant performance and management costs. In this paper, we propose a single system called Pangea that can manage all data—both intermediate and long-lived data, and their buffer/caching, page replacement, data placement optimization, and failure recovery—all in one monolithic distributed storage system, without any layering. We present a detailed performance evaluation of Pangea and show that its performance compares favorably with several widely used layered systems such as Spark.

Query Execution Optimization in Spark SQL

Spark SQL is a big data processing tool for structured data query and analysis. However, due to the execution of Spark SQL, there are multiple times to write intermediate data to the disk, which reduces the execution efficiency of Spark SQL. Targeting on the existing issues, we design and implement an intermediate data cache layer between the underlying file system and the upper Spark core to reduce the cost of random disk I/O. By using the query pre-analysis module, we can dynamically adjust the capacity of cache layer for different queries. And the allocation module can allocate proper memory for each node in cluster. According to the sharing of the intermediate data in the Spark SQL workflow, this paper proposes a cost-based correlation merging algorithm, which can effectively reduce the cost of reading and writing redundant data. This paper develops the SSO (Spark SQL Optimizer) module and integrates it into the original Spark system to achieve the above functions. This paper compares the query performance with the existing Spark SQL by experiment data generated by TPC-H tool. The experimental results show that the SSO module can effectively improve the query efficiency, reduce the disk I/O cost and make full use of the cluster memory resources.

Ideal and Predictable Hit Ratio for Matrix Transposition in Data Caches

Matrix transposition is a fundamental operation, but it may present a very low and hardly predictable data cache hit ratio for large matrices. Safe (worst-case) hit ratio predictability is required in real-time systems. In this paper, we obtain the relations among the cache parameters that guarantee the ideal (predictable) data hit ratio assuming a Least-Recently-Used (LRU) data cache. Considering our analytical assessments, we compare a tiling matrix transposition to a cache oblivious algorithm, modified with phantom padding to improve its data hit ratio. Our results show that, with an adequate tile size, the tiling version results in an equal or better data hit ratio. We also analyze the energy consumption and execution time of matrix transposition on real hardware with pseudo-LRU (PLRU) caches. Our analytical hit/miss assessment enables the usage of a data cache for matrix transposition in real-time systems, since the number of misses in the worst case is bound. In general and high-performance computation, our analysis enables us to restrict the cache resources devoted to matrix transposition with no negative impact, in order to reduce both the energy consumption and the pollution to other computations.

Hierarchical algorithms on hierarchical architectures.

A traditional goal of algorithmic optimality, squeezing out flops, has been superseded by evolution in architecture. Flops no longer serve as a reasonable proxy for all aspects of complexity. Instead, algorithms must now squeeze memory, data transfers, and synchronizations, while extra flops on locally cached data represent only small costs in time and energy. Hierarchically low-rank matrices realize a rarely achieved combination of optimal storage complexity and high-computational intensity for a wide class of formally dense linear operators that arise in applications for which exascale computers are being constructed. They may be regarded as algebraic generalizations of the fast multipole method. Methods based on these hierarchical data structures and their simpler cousins, tile low-rank matrices, are well proportioned for early exascale computer architectures, which are provisioned for high processing power relative to memory capacity and memory bandwidth. They are ushering in a renaissance of computational linear algebra. A challenge is that emerging hardware architecture possesses hierarchies of its own that do not generally align with those of the algorithm. We describe modules of a software toolkit, hierarchical computations on manycore architectures, that illustrate these features and are intended as building blocks of applications, such as matrix-free higher-order methods in optimization and large-scale spatial statistics. Some modules of this open-source project have been adopted in the software libraries of major vendors.This article is part of a discussion meeting issue ‘Numerical algorithms for high-performance computational science’.

RSU-Assisted Adaptive Scheduling for Vehicle-to-Vehicle Data Sharing in Bidirectional Road Scenarios

This study investigates the synergy between centralized and decentralized (i.e., ad hoc) data scheduling in vehicular ad hoc networks (VANETs) for offloading and balancing the workloads of roadside units (RSU) in bidirectional road scenarios. In the centralized scheduling, an RSU schedules data dissemination using a hybrid of infrastructure-to-vehicle (I2V) and vehicle-to-vehicle (V2V) communications. Specifically, RSUs cooperate with each other by transferring unserved requests, and each RSU schedules the data services based on its locally received requests and the transferred requests. In the decentralized scheduling, vehicles driving in opposite directions share the cached data items via V2V communication when out of the coverage of RSU. The ad hoc scheduling can be benefited from the cooperation among RSUs since the chance for V2V data sharing could be enhanced when considering transferred requests into scheduling at each RSU. We formulate a hybrid of centralized and ad hoc data scheduling (HCA) problem, aiming at best exploiting the synergistic effects of I2V and V2V communication based on centralized data broadcast and ad hoc data sharing. On this basis, we propose an RSU Cooperation-based Adaptive Scheduling (RCAS) algorithm that consists of three mechanisms, including a centralized scheduling mechanism at each RSU, an ad hoc scheduling mechanism for vehicles, and a cluster management mechanism. Finally, we build the simulation model and give a comprehensive performance evaluation, which demonstrates the superiority of the proposed solution under a variety of circumstances.

Automatic Safe Data Reuse Detection for the WCET Analysis of Systems With Data Caches

Worst-case execution time (WCET) analysis of systems with data caches is one of the key challenges in real-time systems. Caches exploit the inherent reuse properties of programs, temporarily storing certain memory contents near the processor, in order that further accesses to such contents do not require costly memory transfers. Current worst-case data cache analysis methods focus on specific cache organizations (LRU, locked, ACDC, etc.). In this article, we analyze data reuse (in the worst case) as a property of the program, and thus independent of the data cache. Our analysis method uses Abstract Interpretation on the compiled program to extract, for each static load/store instruction, a linear expression for the address pattern of its data accesses, according to the Loop Nest Data Reuse Theory. Each data access expression is compared to that of prior (dominant) memory instructions to verify whether it presents a guaranteed reuse. Our proposal manages references to scalars, arrays, and non-linear accesses, provides both temporal and spatial reuse information, and does not require the exploration of explicit data access sequences. As a proof of concept we analyze the TACLeBench benchmark suite, showing that most loads/stores present data reuse, and how compiler optimizations affect it. Using a simple hit/miss estimation on our reuse results, the time devoted to data accesses in the worst case is reduced to 27% compared to an always-miss system, equivalent to a data hit ratio of 81%. With compiler optimization, such time is reduced to 6.5%.

Development of the JUNO Conditions Data Management System

The Jiangmen Underground Neutrino Observatory (JUNO) experiment is designed to determine the neutrino mass hierarchy and precisely measure oscillation parameters with an unprecedented energy resolution of 3% at 1 MeV. It is composed of a 20 kton liquid scintillator central detector equipped with 18000 20-inch PMTs and 25000 3-inch PMTs, a water pool with 2000 20-inch PMTs, and a top tracker. Conditions data, coming from calibration and detector monitoring, are heterogeneous, different type of conditions data has different write rates, data format and data volume. JUNO conditions data management system (JCDMS) is developed to homogeneously treat all these heterogeneous conditions data in order to provide easy management and convenient access with both Restful API and web interfaces, support good scalability and maintenance for long time running. The paper describes the status and development of JCDMS including the data model, workflows, interfaces, data caching and performance of the system.

Caching for Mobile Social Networks with Deep Learning: Twitter Analysis for 2016 U.S. Election

As the rise of the portable devices, people usually access the social media such as Twitter and Facebook through wireless networks. Therefore, data transmission rates significant important to the end users. In this work, we discuss the problem of context-aware data caching in the heterogeneous small cell networks to reduce the service delay and how the device-to-device (D2D) and device-to-infrastructure (D2I) improve the system social welfare. In the data-caching model, we explore three types of cache entities, macro cell base stations, small cell base stations, and end user devices. We propose a long short-term memory (LSTM) deep learning model to perform data analysis and extract information content from the data. By knowing the interest of the data to the cache entities, we can cache the data that will most likely to be requested by the end users to reduce service latency. In simulation, we show our proposed algorithm can efficiently reduce the service latency during 2016 U.S. presidential election where mobile user were urgent to request the election information through wireless networks. Comparing with other mechanisms such as using one-to-many matching algorithm or without D2D communication technology, our proposed algorithm improves significantly on the devices performance and system social welfare.

CMS data access and usage studies at PIC Tier-1 and CIEMAT Tier-2

The current computing models from LHC experiments indicate that much larger resource increases would be required by the HL-LHC era (2026+) than those that technology evolution at a constant budget could bring. Since worldwide budget for computing is not expected to increase, many research activities have emerged to improve the performance of the LHC processing software applications, as well as to propose more efficient resource deployment scenarios and data management techniques, which might reduce this expected increase of resources. The massively increasing amounts of data to be processed leads to enormous challenges for HEP storage systems, networks and the data distribution to end-users. These challenges are particularly important in scenarios in which the LHC data would be distributed from small numbers of centers holding the experiment’s data. Enabling data locality relative to computing tasks via local caches on sites seems a very promising approach to hide transfer latencies while reducing the deployed storage space and number of replicas overall. However, this highly depends on the workflow I/O characteristics and available network across sites. A crucial assessment of how the experiments are accessing and using the storage services deployed in WLCG sites, to evaluate and simulate the benefits for several of the new emerging proposals within WLCG/HSF. Studies on access and usage of storage, data access and popularity studies for the CMS workflows executed in the Spanish Tier-1 (PIC) and Tier-2 (CIEMAT) sites supporting CMS activities are reviewed in this report, based on local and experiment monitoring data, spanning more than one year. This is of relevance for simulation of data caches for end-user analysis data, as well as identifying potential areas for storage savings.