Partition Layer Research Articles

Self-aware collaborative edge inference with embedded devices for IIoT

Edge inference and other compute-intensive industrial Internet of Things (IIoT) applications suffer from a bad quality of experience due to the limited and heterogeneous computing and communication resources of embedded devices. To tackle these issues, we propose a model partitioning-based self-aware collaborative edge inference framework. Specifically, the device can adaptively adjust the local model inference scheme by sensing the available computing and communication resources of surrounding devices. When the inference latency requirement cannot be met by local computation, the model should be partitioned for collaborative computation on other devices to improve the inference efficiency. Furthermore, for two typical IIoT scenarios, i.e., bursting and stacking tasks, the latency-aware and throughput-aware collaborative inference algorithms are designed, respectively. Via jointly optimizing the partition layer and collaborative device selection, the optimal inference efficiency, characterized by minimum inference latency and maximum inference throughput, can be obtained. Finally, the performance of our proposal is validated through extensive simulations and tests conducted on 10 Raspberry Pi 4Bs using popular models. Specifically, in the case of two collaborative devices, our platform reaches up to 92.59% latency reduction for bursting tasks and 16.19× throughput growth for stacking tasks. In addition, the divergence between simulations and tests ranges from 1.64% to 9.56% for bursting tasks and from 3.24% to 11.24% for stacking tasks, which indicates that the theoretical performance analyses are solid. For the general case where the data privacy is not considered and the number of collaborative devices is optimally determined, up to 14.76× throughput speed up and 84.04% latency reduction can be obtained.

MistNet: A superior edge-cloud privacy-preserving training framework with one-shot communication

Classical federated learning methods aggregate decentralized data from different devices into a central location for efficient training. However, these approaches raise significant concerns regarding data privacy and requires extensive communication between edge devices and the cloud, leading to high communication costs. To conquer these challenges, we propose an edge-cloud privacy-preserving model training system that enables the cloud and edge devices to collaboratively perform deep neural network(DNN) training, namely MistNet, which has the following advantages. (1) Different from previous works that required frequent communication between the edge and cloud, MistNet requires only one-shot communication. (2) MistNet introduces the rigorous local differential privacy (LDP) technique to the intermediate feature maps, guaranteeing that the user’s local data and model parameters remain undisclosed. (3) In MistNet, the feature extractor is transferred from pre-trained models that are designed for similar application domains. This feature extractor remains fixed during training, thereby eliminating the requirement to synchronize feature extractors across different devices. Furthermore, we first design an object detection algorithm based on Yolov5 with the MistNet framework. Finally, the extensive experiments results on multiple models and datasets demonstrate that by choosing an appropriate partition layer and privacy budget, MistNet achieves lower communication and faster convergence, as well as acceptable model utility while greatly reducing privacy leakage from the released intermediate features. Our framework is primarily designed for image data and has already been applied to Plato. You can find the source code on GitHub page at https://github.com/TL-System/plato.

Hierarchical Transformer with Spatio-temporal Context Aggregation for Next Point-of-interest Recommendation

Next point-of-interest (POI) recommendation is a critical task in location-based social networks, yet remains challenging due to a high degree of variation and personalization exhibited in user movements. In this work, we explore the latent hierarchical structure composed of multi-granularity short-term structural patterns in user check-in sequences. We propose a Spatio-Temporal context AggRegated Hierarchical Transformer (STAR-HiT) for next POI recommendation, which employs stacked hierarchical encoders to recursively encode the spatio-temporal context and explicitly locate subsequences of different granularities. More specifically, in each encoder, the global attention layer captures the spatio-temporal context of the sequence, while the local attention layer performed within each subsequence enhances subsequence modeling using the local context. The sequence partition layer infers positions and lengths of subsequences from the global context adaptively, such that semantics in subsequences can be well preserved. Finally, the subsequence aggregation layer fuses representations within each subsequence to form the corresponding subsequence representation, thereby generating a new sequence of higher-level granularity. The stacking of hierarchical encoders captures the latent hierarchical structure of the check-in sequence, which is used to predict the next visiting POI. Extensive experiments on three public datasets demonstrate that the proposed model achieves superior performance while providing explanations for recommendations.

RTCoInfer: Real-Time Collaborative CNN Inference for Stream Analytics on Ubiquitous Images

Emerging intelligent applications based on accurate and timely stream analytics require real-time CNN inference of massive data continuously generated at the pervasive end devices. Due to the resource constraints, neither computing locally at end devices nor transmitting to remote servers is competent for computation-intensive CNN inference on large-volume images in real-time. Therefore, Collaborative Inference (CI), which conducts inference sequentially from the local device to the remote server with compressed intermediate inference data, is rapidly promoted. Due to the essential communication in collaboration, the CI efficiency is sensitive to network conditions, and will degrade under the unpredictable network fluctuations in practice, which may cause a severe delay in CI and degrade the responsiveness of stream analytics. For accurate and timely stream analytics in practical fluctuating networks, we present RTCoInfer, the real-time CI framework with run-time transmission adaption considering the network conditions. Specifically, we propose a novel Switchable CNN integrating CNNs with different compression rates on the partition layer for the run-time transmission adjustment, and construct a real-time controller determining the compression rate to maintain the real-time CI for stream analytics. Extensive experiments show that, compared with state-of-the-art methods, RTCoInfer achieves better efficiency and unprecedented resilience in real-time stream analytics.

Reversible Entropy-Driven Defect Migration and Insulator-Metal Transition Suppression in VO2 Nanostructures for Phase-Change Electronic Switching.

Oxygen defects are among essential issues and required to be manipulated in correlated electronic oxides with insulator-metal transition (IMT). Besides, surface and interface control are necessary but challenging in field-induced electronic switching towards advanced IMT-triggered transistors and optical modulators. Herein, we demonstrated reversible entropy-driven oxygen defect migrations and reversible IMT suppression in vanadium dioxide (VO2 ) phase-change electronic switching. The initial IMT was suppressed with oxygen defects, which is caused by the entropy change during reversed surface oxygen ionosorption on the VO2 nanostructures. This IMT suppression is reversible and reverts when the adsorbed oxygen extracts electrons from the surface and heals defects again. The reversible IMT suppression observed in the VO2 nanobeam with M2 phase is accompanied by large variations in the IMT temperature. We also achieved irreversible and stable IMT by exploiting an Al2 O3 partition layer prepared by atomic layer deposition (ALD) to disrupt the entropy-driven defect migration. We expected that such reversible modulations would be helpful for understanding the origin of surface-driven IMT in correlated vanadium oxides, and constructing functional phase-change electronic and optical devices.

Phase-Inversion Induced 3D Electrode for Direct Acidic Electroreduction CO2 to Formic acid.

Direct electrochemical CO2 reduction to formic acid (FA) instead of formate is a challenging task due to the high acidity of FA and competitive hydrogen evolution reaction. Herein, 3D porous electrode (TDPE) is prepared by a simple phase inversion method, which can electrochemically reduce CO2 to FA in acidic conditions. Owing to interconnected channels, high porosity, and appropriate wettability, TDPE not only improves mass transport, but also realizes pH gradient to build higher local pH micro-environment under acidic conditions for CO2 reduction compared with planar electrode and gas diffusion electrode. Kinetic isotopic effect experiments demonstrate that the proton transfer becomes the rate-determining step at the pH of 1.8; however, not significant in neutral solution, suggesting that the proton is aiding the overall kinetics. Maximum FA Faradaic efficiency of 89.2% has been reached at pH 2.7 in a flow cell, generating FA concentration of 0.1m. Integrating catalyst and gas-liquid partition layer into a single electrode structure by phase inversion method paves a facile avenue for direct production of FA by electrochemical CO2 reduction.

Determination of pressure difference coefficient of shuttle elevator doors in super high-rise buildings under stack effect

The stack-induced airflow of super high-rise buildings in winter results in obvious indoor environmental problems at the shuttle elevator shaft. Strong pressure difference can cause malfunction of elevators. The estimation of pressure differences at the shuttle elevator doors under different building layout and climate conditions is still a challenge to obtain in engineering application. Considering the difficulties of measuring and simulating the air leakage characteristics of various components, a dimensionless metric, the Pressure Difference Coefficient of Elevator Doors, Cele, is defined according to the theory of pressure distribution in buildings. First, the accuracy of the index calculation results was verified in an elevator test tower. The hand-calculated method to predict the pressure differences at the shuttle elevator doors is convenient and reliable when compared with measurement and simulation. Then, two influence factors on Cele, floor plan and flow exponents are discussed. The determination method of recommended Cele is proposed by analyzing with different building partition layers. The recommended values are given under different shaft heights and temperature difference. Moreover, the novel method and metric were applied in an actual building to determine the number and air tightness of partition layers under the target conditions. This paper provides an evaluation reference for the convenient and effective design of elevator type selection and building layout of super high-rise buildings under stack effect.

Parkinson’s disease progression: Increasing expression of an invariant common core subnetwork

Notable success has been achieved in the study of neurodegenerative conditions using reduction techniques such as principal component analysis (PCA) and sparse inverse covariance estimation (SICE) in positron emission tomography (PET) data despite their widely differing approach. In a recent study of SICE applied to metabolic scans from Parkinson’s disease (PD) patients, we showed that by using PCA to prespecify disease-related partition layers, we were able to optimize maps of functional metabolic connectivity within the relevant networks. Here, we show the potential of SICE, enhanced by disease-specific subnetwork partitions, to identify key regional hubs and their connections, and track their associations in PD patients with increasing disease duration. This approach enabled the identification of a core zone that included elements of the striatum, pons, cerebellar vermis, and parietal cortex and provided a deeper understanding of progressive changes in their connectivity. This subnetwork constituted a robust invariant disease feature that was unrelated to phenotype. Mean expression levels for this subnetwork increased steadily in a group of 70 PD patients spanning a range of symptom durations between 1 and 21 years. The findings were confirmed in a validation sample of 69 patients with up to 32 years of symptoms. The common core elements represent possible targets for disease modification, while their connections to external regions may be better suited for symptomatic treatment.

Urban Road Network Partitioning Based on Bi-Modal Traffic Flows With Multiobjective Optimization

The recent extension of a macroscopic fundamental diagram (MFD) into a bi-modal MFD (or 3D-MFD) provides the relationship among the total network circulating flows and the accumulations of private vehicles and public buses. 3D-MFD reveals the contribution of large occupancy vehicles such as buses in improving urban transportation efficiency. A lot of bi-modal traffic management techniques are introduced based on 3D-MFD to improve the urban traffic efficiency without using detailed origin-destination (OD) information. However, similar to MFD, 3D-MFD is also highly affected by the heterogeneity of a road network. In order to form 3D-MFDs with low scatter to be utilized for further bi-modal traffic management, this paper proposes a partition method to cluster road links into several homogeneous regions for a bi-modal urban network. It is comprised of three layers named as initial partition, merging, and boundary adjusting. At the initial partition layer, Seeded Region Growing (SRG) and the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) are integrated to obtain a number of subregions. A modified Genetic Algorithm (GA) is developed to merge the subregions into larger regions at the merging layer. Then, boundary adjusting is performed by changing the region to which a boundary is clustered to optimize the result. Multi-sensor data collected from Shenzhen in China are utilized to verify the effectiveness of the proposed partition method.

Double exceptional points in grating coupled metal-insulator-metal heterostructure.

In this work we theoretically study the exceptional points and reflection spectra characteristics of a grating coupled metal-insulator-metal heterostructure, which is a non-Hermitian system. Our results show that by selecting suitable geometrical parameters with grating periodicity @150 nm, that satisfy zero reflection condition, double exceptional points appear in a mode bifurcation regime. Furthermore, the thickness of partition metal layer between two cavities plays an important role in controlling the reflection properties of the heterostructure. There is a clear mode splitting when the partition layer allows strong coupling between the two cavity modes. Conversely, in weak coupling regime the mode splitting becomes too close to be distinguished. Moreover, the vanishing of reflection leads to unidirectional reflectionless propagation, which is also known as unidirectional invisibility. With grating periodicity ≥400nm, the transmissions for forward and backward incident directions are no longer the same due to the generation of diffraction. High contrast ratio (≈1) between the two incident directions leads to asymmetric transmission. This work lays the basis for designing double exceptional points and asymmetric transmission in coupled non-Hermitian photonics system. The proposed heterostructure can be a good candidate for new generation optical communications, optical sensing, photo-detection, and nano-photonic devices.

AIgean : An Open Framework for Deploying Machine Learning on Heterogeneous Clusters

AIgean , pronounced like the sea, is an open framework to build and deploy machine learning (ML) algorithms on a heterogeneous cluster of devices (CPUs and FPGAs). We leverage two open source projects: Galapagos , for multi-FPGA deployment, and hls4ml , for generating ML kernels synthesizable using Vivado HLS. AIgean provides a full end-to-end multi-FPGA/CPU implementation of a neural network. The user supplies a high-level neural network description, and our tool flow is responsible for the synthesizing of the individual layers, partitioning layers across different nodes, as well as the bridging and routing required for these layers to communicate. If the user is an expert in a particular domain and would like to tinker with the implementation details of the neural network, we define a flexible implementation stack for ML that includes the layers of Algorithms, Cluster Deployment & Communication, and Hardware. This allows the user to modify specific layers of abstraction without having to worry about components outside of their area of expertise, highlighting the modularity of AIgean . We demonstrate the effectiveness of AIgean with two use cases: an autoencoder, and ResNet-50 running across 10 and 12 FPGAs. AIgean leverages the FPGA’s strength in low-latency computing, as our implementations target batch-1 implementations.

Decoding Multiple Biofunctions of Maca on Its Anti-allergic, Anti-inflammatory, Anti-thrombotic, and Pro-angiogenic Activities.

Four active partition layers and ten isolates, including (5R)- and (5S)-macapyrrolidone A (1a, 1b), and four new alkaloids, (5R)- and (5S)-macapyrrolidone B (2a, 2b) and macapyrrolins D, E (3, 4), were isolated from maca (Lepidium meyenii Walp.), an indigenous food plant from Peru. Derived from the n-hexane layer, the macamide-rich fraction exhibited pro-angiogenic activity on EPC and HUVEC cells. Anti-thrombotic activity was displayed by the polar part of maca extracts (n-butanol and water layers). Both 75% methanol aq. (midlower polar part) and n-hexane (low polar part) layers, which showed signs of fatty acid content, markedly inhibited superoxide and elastase release in an anti-inflammatory assay. The 75% methanol aq. layer showed strong anti-allergic activity, and macapyrrolin A (5) was found active based on β-hexosaminidase release inhibition assays and a ChemGPS-NP experiment. These valuable bioactivity results suggest that maca is a food plant with good benefits for human health.

Systematically assessing genetic strategies for engineering electroactive bacterium to promote bioelectrochemical performances and pollutant removal

Bioelectrochemical system (BES) empowered by electrochemically active microorganisms (EAMs) has a great potential for wastewater treatment. However, their low extracellular electron transfer (EET) capability has restricted the application of BES. In this work, we applied synthetic biology methodology to fundamentally engineer electroactive bacterium for promoted wastewater treatment. At multiple layers, four different genetic engineering strategies, including directly engineering the EET pathways, engineering the bacterial universal messenger, repressing the electron competing reductase, and dynamically managing EET output, were developed and implemented in Shewanella oneidensis, an important model EAM species. The genetic engineered strains improved up to 4.8- and 2.5-fold improvement of current densities over the control in microbial electrosynthesis cell and microbial fuel cell assessments, respectively. Furthermore, using the methyl orange and Cr(VI)-laden wastewaters as the target, the engineered strains exhibited substantially promoted treatment efficiency, achieving up to 18.5- and 5.5-fold over the control. Remarkably, the dynamic control strategy achieved the greatest improvement for wastewater treatment. This study not only provides effective genetic engineering strategies to promote BES performance for wastewater treatment, but also demonstrates the importance of managing EET from the cellular resource partitioning layer for more efficient wastewater treatment.

Enhanced modal matching method for perforated and micro-perforated partitions

Perforated multilayer partitions are widely used in many problems for the aerospace and automotive industries and air-conditioning systems. Combinations of macro and micro-perforated perforations across the constitutive partition layers can provide different physical mechanisms for diverse control strategies depending on the problem requirements. However, there is a lack of unified numerical or analytical description able to provide accurate results over a broad frequency range for a wide range of diameter perforations ranging from supra-millimetric to sub-millimetric apertures. Furthermore, most of them do not account for the beneficial effects on the partition dissipation of the in-hole non-planar modes, albeit evanescent. In this work, an enhanced modal matching (EMM) method is presented that accounts for in-hole high-order modes as well as visco-thermal boundary layer effects inside the holes and over the outer surfaces surrounding the holes. The analytical results have been compared against effective models, numerical models and impedance tube measurements. They show good agreement for single and multi-layer partitions within the corresponding bandwidths of validity. Parametric studies have concluded that the panel thickness-to-hole diameter ratio is a key factor that plays a crucial role on the prominence of the in-hole radial modes and outer visco-thermal effects in the dissipation properties.

Coordinate-based anchor-free module for object detection

Despite the impressive performance of some recent state-of-the-art detectors, small target detection, scale variation, and label ambiguities remain challenges. To tackle these issues, we present a coordinate-based anchor-free (CBAF) module for object detection. It can be used as a branch of a single-shot detector (e.g., RetinaNet or SSD) or predict the output probabilities and coordinates directly. The main idea of the CBAF module is to predict the category and the adjustments to the box of the object by part feature and its contextual part features, which are based on feature maps divided by spatial coordinates. This is inspired by the fact that human beings can infer an entire object by observing the part of the surrounding environment. The CBAF module will encode and decode boxes in the anchor-free manner per feature map with different resolutions during training and testing. During training, we first use the proposed spatial coordinate partition layer to divide feature maps into several parts of size n × n and then propose a contextual building layer to fuse the part and its contextual parts together. We will demonstrate the CBAF module through a concrete implementation. The CBAF module improves AP scores of object detection with nearly no additional computation when working in conjunction with the anchor-based RetinaNet. Furthermore, experimental results on the MS-COCO dataset show that the mAP of the CBAF module has increased by 1.1%, compared with RetinaNet. When the CBAF module works in conjunction with the anchor-based RetinaNet, the mAP increased by 2.2%.

Spectral guided sparse inverse covariance estimation of metabolic networks in Parkinson's disease

In neurodegenerative disorders, a clearer understanding of the underlying aberrant networks facilitates the search for effective therapeutic targets and potential cures. [18F]-fluorodeoxyglucose (FDG) positron emission tomography (PET) imaging data of brain metabolism reflects the distribution of glucose consumption known to be directly related to neural activity. In FDG PET resting-state metabolic data, characteristic disease-related patterns have been identified in group analysis of various neurodegenerative conditions using principal component analysis of multivariate spatial covariance. Notably, among several parkinsonian syndromes, the identified Parkinson's disease-related pattern (PDRP) has been repeatedly validated as an imaging biomarker of PD in independent groups worldwide. Although the primary nodal associations of this network are known, its connectivity is not fully understood. Here, we describe a novel approach to elucidate functional principal component (PC) network connections by performing graph theoretical sparse network derivation directly within the disease relevant PC partition layer of the whole brain data rather than by searching for associations retrospectively in whole brain sparse representations. Using sparse inverse covariance estimation of each overlapping PC partition layer separately, a single coherent network is detected for each layer in contrast to more spatially modular segmentation in whole brain data analysis. Using this approach, the major nodal hubs of the PD disease network are identified and their characteristic functional pathways are clearly distinguished within the basal ganglia, midbrain and parietal areas. Network associations are further clarified using Laplacian spectral analysis of the adjacency matrices. In addition, the innate discriminative capacity of the eigenvector centrality of the graph derived networks in differentiating PD versus healthy external data provides evidence of their validity.

Interval Type-2 Fuzzy C-Means Data Placement Optimization in Scientific Cloud Workflow applications

Scientific workflows stand as practical solutions useful for maintaining data intensive applications representation and execution purposes, which entail not only powerful computing resources, but also massive storage. With the emergence of cloud environment, which enhanced the execution of such applications, the study of workflow placement strategies, as targeted to effectively reduce data movements across data centers, has grown into a highly challenging objective. Given the fact that the workflow execution process is implemented in conformity with a task-execution order, and that each task may deal with either a single or multi-dataset, within a unique data center, various data partitioning or clustering methods have been devised in a bid to retrieve the most optimally effective workflow datasets’ distribution among data centers with the aim of remarkably reducing the datasets’ movements. In this work, a fuzzy data-dependencies based partitioning layer is implemented. More specifically, a dynamic massive data placement strategy is advanced through application of an Interval Type-2 Fuzzy C-Means technique. The latter is opted for as a means whereby the cluster related data centers can be rendered more consistent, thereby, making datasets rather closely associated in terms of related dependency, which helps in remarkably influencing the amounts of transferred data. The proposed strategy is evaluated by means of a simulation technique, using both random and real-world scientific workflows. The performed experiments appear to reveal well that our suggested strategy proves to outperform noticeably the relevant state-of-the-art methods, in that it noticeably helps in reducing the number of data movements across data centers.

P.513 Metabolic syndrome in schizophrenia and polymorphic variants of CYP1A2 and GSTP1 genes

Scientific workflows stand as practical solutions useful for maintaining data intensive applications representation and execution purposes, which entail not only powerful computing resources, but also massive storage. With the emergence of cloud environment, which enhanced the execution of such applications, the study of workflow placement strategies, as targeted to effectively reduce data movements across data centers, has grown into a highly challenging objective. Given the fact that the workflow execution process is implemented in conformity with a task-execution order, and that each task may deal with either a single or multi-dataset, within a unique data center, various data partitioning or clustering methods have been devised in a bid to retrieve the most optimally effective workflow datasets’ distribution among data centers with the aim of remarkably reducing the datasets’ movements. In this work, a fuzzy data-dependencies based partitioning layer is implemented. More specifically, a dynamic massive data placement strategy is advanced through application of an Interval Type-2 Fuzzy C-Means technique. The latter is opted for as a means whereby the cluster related data centers can be rendered more consistent, thereby, making datasets rather closely associated in terms of related dependency, which helps in remarkably influencing the amounts of transferred data. The proposed strategy is evaluated by means of a simulation technique, using both random and real-world scientific workflows. The performed experiments appear to reveal well that our suggested strategy proves to outperform noticeably the relevant state-of-the-art methods, in that it noticeably helps in reducing the number of data movements across data centers.

Collector System Topology Design for Offshore Wind Farm's Repowering and Expansion

After decades of operations, the first batches of offshore wind farms (OWF) are approaching the end of service lifetimes, while the demand for energy transformation is continuously pushing countries around the globe to construct larger OWFs. Therefore, the old OWF's repowering and scale-expansion becomes a possibly-feasible plan which deserves more researches. This paper firstly analyzes the potential benefits of combining the repowering and expansion parts in collector system topology design, which is followed by further analysis and mathematical verifications of economy and reliability advantages for applying cross-substation incorporation (CSI) to the collector system. To achieve the optimal topology design for OWF repowering and expansion, this paper also proposes a multi-layer optimization framework that consists of an offshore substation (OS) refinement layer, an OWF partition layer, and an intra-zone cable connection layer. The whole optimization concerns the wind turbine (WT) capacity deviation between the repowered part and the expanded part, and the final output is generated with the aim of achieving optimal economy and reliability. The proposed topology is tested and compared with traditional radial and ring topologies on a benchmark collector system, and the results verify the high performance in the economy (both short-term and long-term) and reliability.

SparkNN: A Distributed In-Memory Data Partitioning for KNN Queries on Big Spatial Data

The increase in GPS-enabled devices and proliferation of location-based applications have resulted in an abundance of geotagged (spatial) data. As a consequence, numerous applications have emerged that utilize the spatial data to provide different types of location-based services. However, the huge amount of available spatial data presents a challenge to the efficiency of these location-based services. Although the advent of big data frameworks like Apache Spark has enabled the processing of large amounts of data efficiently, they are designed for general (non-spatial) data. That is due to the build-in data partitioning mechanism that does not take into account the spatial proximity of the data. Therefore, these big data frameworks cannot be readily used for spatial analytics such as efficiently answering spatial queries. To fill this gap, this paper proposes SparkNN, an in-memory partitioning and indexing system for answering spatial queries, such as K-nearest neighbor, on big spatial data. SparkNN is implemented on top of Apache Spark and consists of three layers to facilitate efficient spatial queries. The first layer is a spatial-aware partitioning layer, which partitions the spatial data into several partitions ensuring that the load of the partitions is balanced and data objects with close proximity are placed in the same, or neighboring, partitions. The second layer is a local indexing layer, which provides a spatial index inside each partition to speed up the data search within the partition. The third layer is a global index, which is placed in the master node of Spark to route spatial queries to the relevant partitions. The efficiency of SparkNN was evaluated by extensive experiments with big spatial datasets. The results show SparkNN significantly outperforms the state-of-the-art Spark system when evaluated on the same set of queries.