As the size and complexity of high-performance computing (HPC) systems keep growing, scientists' ability to trust the data produced is paramount due to potential data corruption for various reasons, which may stay undetected. While employing machine learning-based anomaly detection techniques could relieve scientists of such concern, it is practically infeasible due to the need for labels for volumes of scientific datasets and the unwanted extra overhead associated. In this paper, we exploit spatial sparsity profiles exhibited in scientific datasets and propose an approach to detect anomalies effectively. Our method first extracts block-level sparse representations of original datasets in the transformed domain. Then it learns from the extracted sparse representations and builds the boundary threshold between normal and abnormal without relying on labeled data. Experiments using real-world scientific datasets show that the proposed approach requires 13% on average (less than 10% in most cases and as low as 0.3%) of the entire dataset to achieve competitive detection accuracy (70.74%-100.0%) as compared to two state-of-the-art unsupervised techniques.
{"title":"Anomaly Detection in Scientific Datasets using Sparse Representation","authors":"Aekyeung Moon, Minjun Kim, Jiaxi Chen, S. Son","doi":"10.1145/3588982.3603610","DOIUrl":"https://doi.org/10.1145/3588982.3603610","url":null,"abstract":"As the size and complexity of high-performance computing (HPC) systems keep growing, scientists' ability to trust the data produced is paramount due to potential data corruption for various reasons, which may stay undetected. While employing machine learning-based anomaly detection techniques could relieve scientists of such concern, it is practically infeasible due to the need for labels for volumes of scientific datasets and the unwanted extra overhead associated. In this paper, we exploit spatial sparsity profiles exhibited in scientific datasets and propose an approach to detect anomalies effectively. Our method first extracts block-level sparse representations of original datasets in the transformed domain. Then it learns from the extracted sparse representations and builds the boundary threshold between normal and abnormal without relying on labeled data. Experiments using real-world scientific datasets show that the proposed approach requires 13% on average (less than 10% in most cases and as low as 0.3%) of the entire dataset to achieve competitive detection accuracy (70.74%-100.0%) as compared to two state-of-the-art unsupervised techniques.","PeriodicalId":432974,"journal":{"name":"Proceedings of the First Workshop on AI for Systems","volume":"32 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-08-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129749033","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Burak Aksar, Efe Sencan, B. Schwaller, V. Leung, Jim Brandt, B. Kulis, Manuel Egele, A. Coskun
Supercomputers are highly sophisticated computing systems designed to handle complex and computationally intensive tasks. Despite their tremendous efficiency, performance problems still arise due to various factors, such as load imbalance, network congestion, and software-related issues. Monitoring frameworks are commonly used to collect telemetry data, which helps identify potential issues before they become critical or debug problems. However, telemetry analytics is essentially a big data problem that is becoming increasingly difficult to manage due to terabytes of telemetry data collected daily. Owing to the limitations of manual analysis, recent analytics frameworks leverage automated machine learning (ML)-based frameworks to identify patterns and anomalies in this data, enabling system administrators and users to take appropriate action towards resolving performance problems quickly. This paper explores the benefits and challenges of ML-based frameworks that automate performance diagnostics, particularly focusing on labeled training data requirements and deployment challenges. We argue that ML-based frameworks can achieve desirable performance diagnosis results while reducing the need for large labeled data sets, and we demonstrate successful prototypes that are suitable for rapid deployment on real-world systems.
{"title":"Towards Practical Machine Learning Frameworks for Performance Diagnostics in Supercomputers","authors":"Burak Aksar, Efe Sencan, B. Schwaller, V. Leung, Jim Brandt, B. Kulis, Manuel Egele, A. Coskun","doi":"10.1145/3588982.3603609","DOIUrl":"https://doi.org/10.1145/3588982.3603609","url":null,"abstract":"Supercomputers are highly sophisticated computing systems designed to handle complex and computationally intensive tasks. Despite their tremendous efficiency, performance problems still arise due to various factors, such as load imbalance, network congestion, and software-related issues. Monitoring frameworks are commonly used to collect telemetry data, which helps identify potential issues before they become critical or debug problems. However, telemetry analytics is essentially a big data problem that is becoming increasingly difficult to manage due to terabytes of telemetry data collected daily. Owing to the limitations of manual analysis, recent analytics frameworks leverage automated machine learning (ML)-based frameworks to identify patterns and anomalies in this data, enabling system administrators and users to take appropriate action towards resolving performance problems quickly. This paper explores the benefits and challenges of ML-based frameworks that automate performance diagnostics, particularly focusing on labeled training data requirements and deployment challenges. We argue that ML-based frameworks can achieve desirable performance diagnosis results while reducing the need for large labeled data sets, and we demonstrate successful prototypes that are suitable for rapid deployment on real-world systems.","PeriodicalId":432974,"journal":{"name":"Proceedings of the First Workshop on AI for Systems","volume":"83 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-08-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125756576","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Edson Ramiro Lucas Filho, Lun Yang, Kebo Fu, H. Herodotou
Modern data storage systems optimize data access by distributing data across multiple storage tiers and caches, based on numerous tiering and caching policies. The policies' decisions, and in particular the ones related to data prefetching, can severely impact the performance of the entire storage system. In recent years, various machine learning algorithms have been employed to model access patterns in complex data storage workloads. Even though data storage systems handle a constantly changing stream of file requests, current approaches continue to train their models offline in a batch-based approach. In this paper, we investigate the use of streaming machine learning to support data prefetching decisions in data storage systems as it introduces various advantages such as high training efficiency, high prediction accuracy, and high adaptability to changing workload patterns. After extracting a representative set of features in an online fashion, streaming machine learning models can be trained and tested while the system is running. To validate our methodology, we present one streaming classification model to predict the next file offset to be read in a file. We assess the model's performance using production traces provided by Huawei Technologies and demonstrate that streaming machine learning is a feasible approach with low memory consumption and minimal training delay, facilitating accurate predictions in real-time.
{"title":"Streaming Machine Learning for Supporting Data Prefetching in Modern Data Storage Systems","authors":"Edson Ramiro Lucas Filho, Lun Yang, Kebo Fu, H. Herodotou","doi":"10.1145/3588982.3603608","DOIUrl":"https://doi.org/10.1145/3588982.3603608","url":null,"abstract":"Modern data storage systems optimize data access by distributing data across multiple storage tiers and caches, based on numerous tiering and caching policies. The policies' decisions, and in particular the ones related to data prefetching, can severely impact the performance of the entire storage system. In recent years, various machine learning algorithms have been employed to model access patterns in complex data storage workloads. Even though data storage systems handle a constantly changing stream of file requests, current approaches continue to train their models offline in a batch-based approach. In this paper, we investigate the use of streaming machine learning to support data prefetching decisions in data storage systems as it introduces various advantages such as high training efficiency, high prediction accuracy, and high adaptability to changing workload patterns. After extracting a representative set of features in an online fashion, streaming machine learning models can be trained and tested while the system is running. To validate our methodology, we present one streaming classification model to predict the next file offset to be read in a file. We assess the model's performance using production traces provided by Huawei Technologies and demonstrate that streaming machine learning is a feasible approach with low memory consumption and minimal training delay, facilitating accurate predictions in real-time.","PeriodicalId":432974,"journal":{"name":"Proceedings of the First Workshop on AI for Systems","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-08-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131033095","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Proceedings of the First Workshop on AI for Systems","authors":"","doi":"10.1145/3588982","DOIUrl":"https://doi.org/10.1145/3588982","url":null,"abstract":"","PeriodicalId":432974,"journal":{"name":"Proceedings of the First Workshop on AI for Systems","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122328038","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：