Execution Time Prediction Research Articles

EXPRESS: A Framework for Execution Time Prediction of Concurrent CNNs on Xilinx DPU Accelerator

Deep learning Processor Unit (DPU) is a highly configurable CNN accelerator that supports a variety of CNNs and can be implemented with multiple instances on the same FPGA. Many applications deploy concurrent execution of different CNNs and in such a setting, an execution time predictor can help “optimize” the DPU configurations to meet the performance requirements of different tasks. We characterize CNN execution on DPUs and reduce the variability in execution time due to interference from the operating system. Subsequently, we propose a machine learning-based framework (EXPRESS) to predict the execution time of any given CNN on a DPU configuration, considering CNN, DPU, and bus characteristics. We improvise EXPRESS to support heterogeneous CNNs in EXPRESS-2.0 by making features independent of the number of CNNs. Our entire experimentation is based on data from a real FPGA board for 16 standard CNNs. Our frameworks, EXPRESS and EXPRESS-2.0, significantly outperform state-of-the-art by achieving an average execution time prediction error of 2.2% and 0.7%, respectively. We illustrate the effectiveness of this low prediction error for design space exploration, which is very useful for embedded system application developers.

Analyzing Machine Learning Models for Activity Recognition Using Homomorphically Encrypted Real-World Smart Home Datasets: A Case Study

The era of digitization and IoT devices is marked by the constant storage of massive amounts of data. The growing adoption of smart home environments, which use sensors and devices to monitor and control various aspects of daily life, underscores the need for effective privacy and security measures. HE is a technology that enables computations on encrypted data, preserving confidentiality. As a result, researchers have developed methodologies to protect user information, and HE is one of the technologies that make it possible to perform computations directly on encrypted data and produce results using this encrypted information. Thus, this research study compares the performance of three ML models, XGBoost, Random Forest, and Decision Classifier, on a real-world smart home dataset using both with and without FHE. Practical results demonstrate that the Decision Classifier showed remarkable results, maintaining high accuracy with FHE and even surpassing its plaintext performance, suggesting that encryption can enhance model accuracy under certain conditions. Additionally, Random Forest showed efficiency in terms of execution time and low prediction errors with FHE, making it a strong candidate for encrypted data processing in smart homes. These findings highlight the potential of FHE to set new privacy standards, advancing secure and privacy-preserving technologies in smart environments.

A novel query execution time prediction approach based on operator iterate-aware of the execution plan on the graph database

Query execution time prediction is essential for database query optimization tasks, such as query scheduling, progress monitoring, and resource allocation. In the query execution time prediction tasks, the query plan is often used as the modeling object of a prediction model. Although the learning-based prediction models have been proposed to capture plan features, there are two limitations need to be considered more. First, the parent–child dependencies between plan operators can be captured, but the operator’s branch independence cannot be distinguished. Second, each operator’s output row is its following operator input, but the data iterate transfer operations between operators are ignored. In this study, we propose a graph query execution time prediction model containing a plan module, a query module, a plan-query module, and a prediction module to improve prediction effectiveness. Specifically, the plan module is used to capture the data iterate transfer operations and distinguish independent of branch operators; the query module is used to learn features of query terms that have an influence on the composition of operators; the plan-query interaction module is used to learn the logical correlations of plan and query. The experiment on datasets proves the effectiveness of the operator iterate-aware and query-plan interaction method in our proposed graph query execution prediction model.

CAMP: a hierarchical cache architecture for multi-core mixed criticality processors

CAMP proposes a hierarchical cache subsystem for multi-core mixed criticality processors, focusing on ensuring worst-case execution time (WCET) predictability in automotive applications. It incorporates criticality-aware locked L1 and L2 caches, reconfigurable at mode change intervals, along with criticality-aware last level cache partitioning. Evaluation using CACOSIM, Moola Multicore simulator, and CACTI simulation tools confirms the suitability of CAMP for keeping high-criticality jobs within timing budgets. A practical case study involving an automotive wake-up controller using the sniper v7.2 architecture simulator further validates its usability in real-world mixed criticality applications. CAMP presents a promising cache architecture for optimized multi-core mixed criticality systems.

A Novel Multi-Task Performance Prediction Model for Spark

Performance prediction of Spark plays a vital role in cluster resource management and system efficiency improvement. The performance of Spark is affected by several variables, such as the size of the input data, the computational power of the system, and the complexity of the algorithm. At the same time, less research has focused on multi-task performance prediction models for Spark. To address these challenges, we propose a multi-task Spark performance prediction model. The model integrates a multi-head attention mechanism and a convolutional neural network. It implements the prediction of execution times for single or multiple Spark applications. Firstly, the data are dimensionally reduced by a dimensionality reduction algorithm and fed into the model. Secondly, the model integrates a multi-head attention mechanism and a convolutional neural network. It captures complex relationships between data features and uses these features for Spark performance prediction. Finally, we use residual connections to prevent overfitting. To validate the performance of the model, we conducted experiments on four Spark benchmark applications. Compared to the benchmark prediction model, our model obtains better performance metrics. In addition, our model predicts multiple Spark benchmark applications simultaneously and maintains deviations within permissible limits. It provides a novel way for the assessment and optimization of Spark.

Improving prediction of computational job execution times with machine learning

SummaryPredicting resource consumption and run time of computational workloads is crucial for efficient resource allocation, or cost and energy optimization. In this paper, we evaluate various machine learning techniques to predict the execution time of computational jobs. For experiments we use datasets from two application areas: scientific workflow management and data processing in the ALICE experiment at CERN. We apply a two‐stage prediction method and evaluate its performance. Other evaluated aspects include: (1) comparing performance of global (per‐workflow) versus specialized (per‐job) models; (2) impact of prediction granularity in the first stage of the two‐stage method; (3) using various feature sets, feature selection, and feature importance analysis; (4) applying symbolic regression in addition to classical regressors. Our results provide new valuable insights on using machine learning techniques to predict the runtime behavior of computational jobs.

Cost‐ and performance‐aware resource selection for parallel software on heterogeneous cloud

SummaryCloud providers offer flexible infrastructures and on‐demand services, including the capability to deploy low cost virtual resources of many different types. However, the diversity of cloud resources followed by the important trade‐off between cost and performance makes the resource selection a challenging task for users in the case of parallel communication‐intensive software. The paper presents cost‐ and performance‐aware resource selection for parallel discrete element method (DEM) software as a service (SaaS) on heterogeneous OpenStack cloud. The developed resource selection uses preliminary application‐specific benchmarks of size smaller than targeted problems and the performance prediction based on speedup of parallel computations to obtain Pareto optimal solutions and to select the best configuration of containers from user's perspective. Hybrid parallelization of DEM software is developed by using OpenCL for shared‐memory multi‐core architectures and MPI for internode communications on distributed‐memory computer clusters. Round up and proportional pricing schemes are examined and compared from a user's perspective. Lower cost of computations obtained by using the proportional pricing scheme is always preferable for users. However, the difference approaches 1.0% of the cost calculated by using proportional pricing scheme, when long lasting computations are performed. The prediction tends to underestimate the execution time of DEM SaaS, but its accuracy is sufficient to obtain the same Pareto optimal solutions by using measured and predicted execution times. Pareto front and linear scalarization propose to select configurations of containers capable of exploiting higher memory bandwidth, which is specific to memory bandwidth bound DEM computations.

Optimizing job scheduling by using broad learning to predict execution times on HPC clusters

Optimizing job scheduling by using broad learning to predict execution times on HPC clusters

AFMC: An alignment framework for multiple computing services and providers

SummaryThe Hirschberg algorithm is commonly used for protein sequence alignment, which is a very important task in bioinformatics. This article presents the AFMC framework for using the Hirschberg method to perform sequence alignment in multiple cloud computing services of different models, such as Infrastructure‐as‐a‐Service and Function‐as‐a‐Service (FaaS). Experiments were carried out in which several instances of AWS EC2, Azure VMs and Google Compute Engine as well as varied configurations of AWS Lambda, Azure Function, and Google Cloud Function were used to pairwise align COVID‐19 spike proteins. The services were submitted to different levels of simultaneity to align the genetic sequences. The findings reveal that there is a tradeoff between predicted execution time and cost for this application, for example, FaaS‐oriented cloud service models generally took less time to process the workloads. On the other hand, it was observed that, as the level of concurrence increased, there was a marked augmentation in cost. In this context, a framework that provides multi cloud solutions for bioinformatics such as AFMC is essential.

EXSCALATE: An Extreme-Scale Virtual Screening Platform for Drug Discovery Targeting Polypharmacology to Fight SARS-CoV-2

The social and economic impact of the COVID-19 pandemic demands a reduction of the time required to find a therapeutic cure. In this paper, we describe the EXSCALATE molecular docking platform capable to scale on an entire modern supercomputer for supporting extreme-scale virtual screening campaigns. Such virtual experiments can provide in short time information on which molecules to consider in the next stages of the drug discovery pipeline, and it is a key asset in case of a pandemic. The EXSCALATE platform has been designed to benefit from heterogeneous computation nodes and to reduce scaling issues. In particular, we maximized the accelerators’ usage, minimized the communications between nodes, and aggregated the I/O requests to serve them more efficiently. Moreover, we balanced the computation across the nodes by designing an ad-hoc workflow based on the execution time prediction of each molecule. We deployed the platform on two HPC supercomputers, with a combined computational power of 81 PFLOPS, to evaluate the interaction between 70 billion of small molecules and 15 binding-sites of 12 viral proteins of SARS-CoV-2. The experiment lasted 60 hours and it performed more than one trillion ligand-pocket evaluations, setting a new record on the virtual screening scale.

SSO-RBNN driven brain tumor classification with Saliency-K-means segmentation technique

Early-stage diagnosis of Brain Tumor leads to better chance of cure from this deadliest disease across the globe. Existing schemes on brain tumor classification use machine learning, convolutional neural networks, Generative Adversarial Networks, and deep learning schemes. However, more execution time and uncertain predictions leading additional process to cross-check the obtained results. In this paper, a classification model Saliency-K-mean-SSO-RBNN is formulated including a new hybrid salience-K-mean segmentation technique along with utilizing the advantage of social spider optimization (SSO) algorithm in Radial Basis Neural Network (RBNN). Hybrid Saliency Map with K-means cluster-based segmentation approach is formulated to segment the tumor region. As Saliency map spotlights on eye catching region within target image, segmented image is fetched to feature extraction phase by considering multiresolution wavelet transform, Principal Component, Kurtosis, Skewness, Inverse Difference Moment (IDM), Cosine transform. Feature vector is then processed for an efficient classification using RBNN by optimizing the cluster center through SSO. RBNN with Gaussian kernel depicts a low complex model for classification. Saliency-K-mean-SSO-RBNN and new hybrid Saliency-K-mean segmentation are validated on standard datasets and compared with existing schemes with regard to specificity, precision, sensitivity, F1 score, MCC, Kappa coefficient and complexity. Saliency-K-mean-SSO-RBNN, yielding a classification accuracy in three datasets as 96%, 92%, and 94%.

Evaluating execution time predictions on GPU kernels using an analytical model and machine learning techniques

Predicting the performance of applications executed on GPUs is a great challenge and is essential for efficient job schedulers. There are different approaches to do this, namely analytical modeling and machine learning (ML) techniques. Machine learning requires large training sets and reliable features, nevertheless it can capture the interactions between architecture and software without manual intervention.In this paper, we compared a BSP-based analytical model to predict the time of execution of kernels executed over GPUs. The comparison was made using three different ML techniques. The analytical model is based on the number of computations and memory accesses of the GPU, with additional information on cache usage obtained from profiling. The ML techniques Linear Regression, Support Vector Machine, and Random Forest were evaluated over two scenarios: first, data input or features for ML techniques were the same as the analytical model and, second, using a process of feature extraction, which used correlation analysis and hierarchical clustering. Our experiments were conducted with 20 CUDA kernels, 11 of which belonged to 6 real-world applications of the Rodinia benchmark suite, and the other were classical matrix-vector applications commonly used for benchmarking. We collected data over 9 NVIDIA GPUs in different machines.We show that the analytical model performs better at predicting when applications scale regularly. For the analytical model a single parameter λ is capable of adjusting the predictions, minimizing the complex analysis in the applications. We show also that ML techniques obtained high accuracy when a process of feature extraction is implemented. Sets of 5 and 10 features were tested in two different ways, for unknown GPUs and for unknown Kernels. For ML experiments with a process of feature extractions, we got errors around 1.54% and 2.71%, for unknown GPUs and for unknown Kernels, respectively.

Cost-Based or Learning-Based?

Traditional cost-based optimizers are efficient and stable to generate optimal plans for simple SQL queries, but they may not generate high-quality plans for complicated queries. Thus learning-based optimizers have been proposed recently that can learn high-quality plans based on past experiences. However, learning-based optimizers cannot work well for dynamic workloads that have different distributions with training examples. In this paper, we propose a hybrid optimizer that adopts the advantages and avoids the shortcomings of these two types of optimizers, which first generates high-quality candidate plans from each type of optimizers and then selects the best plan from the candidates. There are two challenges. (1) How to generate high-quality candidates? We propose a hint-based candidate generation method that leverages the learning-based method to generate highly beneficial hints and then uses a cost-based method to supplement the hints to generate complete plans as candidates. (2) How to evaluate different candidate plans and select the best one? We propose an uncertainty-based optimal plan selection model, which predicts the execution time and the uncertainty for each plan. The uncertainty reflects the confidence of the execution time prediction. We select the plan using the uncertainty model. Experiment results on real datasets showed that our method outperformed the state-of-the-art baselines, and reduced the total latency by 25% and the tail latency by 65% compared to PostgreSQL.

Predicting remaining execution time of business process instances via auto-encoded transition system

As an important task in business process management, remaining time prediction for business process instances has attracted extensive attentions. However, most of the traditional remaining time prediction approaches only take into account formal process models and cannot handle large-scale event logs in an effective manner. Although machine learning and deep learning have been recently applied to the remaining time prediction task, these approaches cannot incorporate domain knowledge naturally. To overcome these weaknesses of existing studies, we propose a remaining execution time prediction approach based on a novel auto-encoded transition system, which can enhance the complementarity of process modeling and deep learning techniques. Through auto-encoding the event-level and state-level features, the proposed approach can represent process instances in a comprehensive and compact form. Furthermore, a transfer learning strategy is proposed to train the remaining time prediction model so as to avoid overfitting and improve the accuracy of prediction. We conduct extensive experiments on four real-world datasets to verify the effectiveness of the proposed approach. The results show its superiority over several state-of-the-art approaches.

A Constructive Heuristic for Automated Parallel Tests Assembly

Parallel tests contain different items but have the same measurement properties. They are administered at the same or different time slots and their measurement results must be comparable. The problem of automated parallel tests assembly is studied for a long time, and many (mostly improvement) heuristic solutions are proposed and elaborated in literature. Such approaches frequently suffer from algorithms of unpredictable execution time, forcing the methods to terminate execution when some time limit or solution quality is reached. This paper proposes an efficient method of polynomial complexity, as a complete solution to the automated parallel tests assembly problem. The method uses the idea of Nawaz, Enscore, and Ham constructive heuristic algorithm to reduce the number of examined permutations, originally exploited for solving the permutation flow-shop sequencing problem. We compared the experimental results of the proposed method with two methods based on improvement heuristics that solve the same problem formulation, simulated annealing and variable neighborhood search. The main advantages of the proposed method are predictable execution time and implementation simplicity. Achieved quality of assembled tests, combined with predictable test assembly execution time, may be of particular interest in cases when computational resources for test assembly and administering are overloaded.

Execution Time Prediction for Cypher Queries in the Neo4j Database Using a Learning Approach

With database management systems becoming complex, predicting the execution time of graph queries before they are executed is one of the challenges for query scheduling, workload management, resource allocation, and progress monitoring. Through the comparison of query performance prediction methods, existing research works have solved such problems in traditional SQL queries, but they cannot be directly applied in Cypher queries on the Neo4j database. Additionally, most query performance prediction methods focus on measuring the relationship between correlation coefficients and retrieval performance. Inspired by machine-learning methods and graph query optimization technologies, we used the RBF neural network as a prediction model to train and predict the execution time of Cypher queries. Meanwhile, the corresponding query pattern features, graph data features, and query plan features were fused together and then used to train our prediction models. Furthermore, we also deployed a monitor node and designed a Cypher query benchmark for the database clusters to obtain the query plan information and native data store. The experimental results of four benchmarks showed that the average mean relative error of the RBF model reached 16.5% in the Northwind dataset, 12% in the FIFA2021 dataset, and 16.25% in the CORD-19 dataset. This experiment proves the effectiveness of our proposed approach on three real-world datasets.

A node2vec-based graph embedding approach for unified assembly process information modeling and workstep execution time prediction

The trend of customized production results in the demand for higher level of automation, in which artificial intelligence decision-making dominates. As the core of smart manufacturing systems, intelligent services stand a significant role in the analysis, prediction, and adjustment of production process, which is inseparable from the effective semantic modeling of the procedures and elements involved. However, there is an absence of unified modeling of assembly process, including both geometric and non-geometric information, leading to the incomprehensiveness when providing data support for intelligent services. To fill this gap, a generic node2vec-based parameterized representation of geometric elements and assembly constraints approach is proposed. Firstly, the information structure of assembly process is established, in which the geometric elements and topological relationships of the product are abstracted into a network. Secondly, node2vec is adopted for the graph embedding to generate preset dimension vectors corresponding to the nodes in the geometric network. As the edges in the network, the vectors corresponding to the assembly constraints, which are regarded as the parameterized representations, can be obtained through node vector calculation. Moreover, an assembly workstep execution time prediction method based on historical data is introduced with the parameterized representations of assembly constraints as the carriers of geometric topological information. At last, an industrial case study is illustrated to show the entire process of constraint parameterized representations and workstep execution time prediction, indicating the feasibility and availability of the method proposed.

Edge Computing Using Embedded Webserver with Mobile Device for Diagnosis and Prediction of Metastasis in Histopathological Images

Diagnosis of different breast cancer stages using histopathology whole slide images is the gold standard in grading the tissue metastasis. Traditional diagnosis involves labor intensive procedures and is prone to human errors. Computer aided diagnosis assists medical experts as a second opinion tool in early detection which prevents further proliferation. Computing facilities have emerged to an extent where algorithms can attain near human accuracy in prediction of diseases, offering better treatment to curb further proliferation. The work introduced in the paper provides an interface in mobile platform, which enables the user to input histopathology image and obtain the prediction results with its class probability through embedded web-server. The trained deep convolutional neural networks model is deployed into a microcomputer-based embedded system after hyper-parameter tuning, offering congruent performance. The implementation results show that the embedded platform with custom-trained CNN model is suitable for medical image classification, as it takes less execution time and mean prediction time. It is also noticed that customized CNN classifier model outperforms pre-trained models when used in embedded platforms for prediction and classification of histopathology images. This work also emphasizes the relevance of portable and flexible embedded device in real time clinical applications.

Execution time prediction model for parallel GPU realizations of discrete transforms computation algorithms

Execution time prediction model for parallel GPU realizations of discrete transforms computation algorithms

Estimating Query Timings in Elasticsearch

In a shared Elasticsearch environment it can be useful to know how long a particular query will take to execute. This information can be used to enforce rate limiting or distribute requests equitably among multiple clusters. Elasticsearch uses multiple Lucene instances on multiple hosts as an underlying search engine implementation, but this abstraction makes it difficult to predict execution with previously known predictors such as the number of postings. This research investigates the ability of different pre-retrieval statistics, available through Elasticsearch, to accurately predict the execution time of queries on a typical Elasticsearch cluster. The number of terms in a query and the Total Term Frequency (TTF) from Elasticsearch’s API are found to significantly predict execution time. Regression models are then built and compared to find the most accurate method for predicting query time.