I. Calciu, M. Imran, Ivan Puddu, Sanidhya Kashyap, H. Maruf, O. Mutlu, Aasheesh Kolli
Disaggregated memory provides many cost savings and resource provisioning benefits for current datacenters, but software systems enabling disaggregated memory access result in high performance penalties. These systems require intrusive code changes to port applications for disaggregated memory or employ slow virtual memory mechanisms to avoid code changes. Such mechanisms result in high overhead page faults to access remote data and high dirty data amplification when tracking changes to cached data at page-granularity. In this paper, we propose a fundamentally new approach for disaggregated memory systems, based on the observation that we can use local cache coherence to track applications' memory accesses transparently, without code changes, at cache-line granularity. This simple idea (1) eliminates page faults from the application critical path when accessing remote data, and (2) decouples the application memory access tracking from the virtual memory page size, enabling cache-line granularity dirty data tracking and eviction. Using this observation, we implemented a new software runtime for disaggregated memory that improves average memory access time and reduces dirty data amplification1.
{"title":"Using Local Cache Coherence for Disaggregated Memory Systems","authors":"I. Calciu, M. Imran, Ivan Puddu, Sanidhya Kashyap, H. Maruf, O. Mutlu, Aasheesh Kolli","doi":"10.1145/3606557.3606561","DOIUrl":"https://doi.org/10.1145/3606557.3606561","url":null,"abstract":"Disaggregated memory provides many cost savings and resource provisioning benefits for current datacenters, but software systems enabling disaggregated memory access result in high performance penalties. These systems require intrusive code changes to port applications for disaggregated memory or employ slow virtual memory mechanisms to avoid code changes. Such mechanisms result in high overhead page faults to access remote data and high dirty data amplification when tracking changes to cached data at page-granularity. In this paper, we propose a fundamentally new approach for disaggregated memory systems, based on the observation that we can use local cache coherence to track applications' memory accesses transparently, without code changes, at cache-line granularity. This simple idea (1) eliminates page faults from the application critical path when accessing remote data, and (2) decouples the application memory access tracking from the virtual memory page size, enabling cache-line granularity dirty data tracking and eviction. Using this observation, we implemented a new software runtime for disaggregated memory that improves average memory access time and reduces dirty data amplification1.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"57 1","pages":"21 - 28"},"PeriodicalIF":0.0,"publicationDate":"2023-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"45832972","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}
Resource disaggregation is an approach to separate different hardware resources into independent pools in a data center, so that these pools can be easily managed and their resources can be allocated in a tight but unbounded way. The past decade has seen research and practices in realizing the resource-disaggregation idea on regular servers. We advocate for a physically disaggregated data center, where disaggregated resource pools consist of hardware devices, not servers. Physical disaggregation could unlock another level of benefits in resource disaggregation, including further improved cost saving, easier maintenance and scaling, and more customization. This paper presents our efforts in building an end-to-end physically disaggregated data center, including the design and implementation of disaggregated hardware devices, networking systems for connecting these devices, operating systems for orchestrating them, and porting of traditional and cloud-computing applications to this physically disaggregated platform.
{"title":"Make It Real: An End-to-End Implementation of A Physically Disaggregated Data Center","authors":"Yiying Zhang","doi":"10.1145/3606557.3606559","DOIUrl":"https://doi.org/10.1145/3606557.3606559","url":null,"abstract":"Resource disaggregation is an approach to separate different hardware resources into independent pools in a data center, so that these pools can be easily managed and their resources can be allocated in a tight but unbounded way. The past decade has seen research and practices in realizing the resource-disaggregation idea on regular servers. We advocate for a physically disaggregated data center, where disaggregated resource pools consist of hardware devices, not servers. Physical disaggregation could unlock another level of benefits in resource disaggregation, including further improved cost saving, easier maintenance and scaling, and more customization. This paper presents our efforts in building an end-to-end physically disaggregated data center, including the design and implementation of disaggregated hardware devices, networking systems for connecting these devices, operating systems for orchestrating them, and porting of traditional and cloud-computing applications to this physically disaggregated platform.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"57 1","pages":"1 - 9"},"PeriodicalIF":0.0,"publicationDate":"2023-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"45983312","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}
Hardware disaggregation has emerged as one of the most fundamental shifts in how we build computer systems over the past decades. While disaggregation has been successful for several types of resources (storage, power, and others), memory disaggregation has yet to happen. We make the case that the time for memory disaggregation has arrived. We look at past successful disaggregation stories and learn that their success depended on two requirements: addressing a burning issue and being technically feasible. We examine memory disaggregation through this lens and find that both requirements are finally met. Once available, memory disaggregation will require software support to be used effectively. We discuss some of the challenges of designing an operating system that can utilize disaggregated memory for itself and its applications.
{"title":"Memory disaggregation: why now and what are the challenges","authors":"M. Aguilera, Emmanuel Amaro, Nadav Amit, Erika Hunhoff, Anil Yelam, Gerd Zellweger","doi":"10.1145/3606557.3606563","DOIUrl":"https://doi.org/10.1145/3606557.3606563","url":null,"abstract":"Hardware disaggregation has emerged as one of the most fundamental shifts in how we build computer systems over the past decades. While disaggregation has been successful for several types of resources (storage, power, and others), memory disaggregation has yet to happen. We make the case that the time for memory disaggregation has arrived. We look at past successful disaggregation stories and learn that their success depended on two requirements: addressing a burning issue and being technically feasible. We examine memory disaggregation through this lens and find that both requirements are finally met. Once available, memory disaggregation will require software support to be used effectively. We discuss some of the challenges of designing an operating system that can utilize disaggregated memory for itself and its applications.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"57 1","pages":"38 - 46"},"PeriodicalIF":0.0,"publicationDate":"2023-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"47625536","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}
Divyanshu Saxena, T. Ji, Arjun Singhvi, Junaid Khalid, Aditya Akella
Navigating the performance and efficiency trade-offs is critical for serverless platforms, where the providers ideally want to give the illusion of warm function startups while maintaining low resource costs. Limited controls, provided via toggling sandboxes between warm and cold states and keepalives, force operators to sacrifice significant resources to achieve good performance.
{"title":"Navigating Performance-Efficiency Tradeoffs in Serverless Computing: Deduplication to the Rescue!","authors":"Divyanshu Saxena, T. Ji, Arjun Singhvi, Junaid Khalid, Aditya Akella","doi":"10.1145/3606557.3606564","DOIUrl":"https://doi.org/10.1145/3606557.3606564","url":null,"abstract":"Navigating the performance and efficiency trade-offs is critical for serverless platforms, where the providers ideally want to give the illusion of warm function startups while maintaining low resource costs. Limited controls, provided via toggling sandboxes between warm and cold states and keepalives, force operators to sacrifice significant resources to achieve good performance.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"57 1","pages":"47 - 53"},"PeriodicalIF":0.0,"publicationDate":"2023-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"43927084","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}
Henrique Fingler, Zhiting Zhu, Esther Yoon, Zhipeng Jia, E. Witchel, C. Rossbach
Serverless platforms have been attracting applications from traditional platforms because infrastructure management responsibilities are shifted from users to providers. Many applications well-suited to serverless environments could leverage GPU acceleration to enhance their performance. Unfortunately, current serverless platforms do not expose GPUs to serverless applications.
{"title":"Disaggregated GPU Acceleration for Serverless Applications","authors":"Henrique Fingler, Zhiting Zhu, Esther Yoon, Zhipeng Jia, E. Witchel, C. Rossbach","doi":"10.1145/3606557.3606560","DOIUrl":"https://doi.org/10.1145/3606557.3606560","url":null,"abstract":"Serverless platforms have been attracting applications from traditional platforms because infrastructure management responsibilities are shifted from users to providers. Many applications well-suited to serverless environments could leverage GPU acceleration to enhance their performance. Unfortunately, current serverless platforms do not expose GPUs to serverless applications.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"57 1","pages":"10 - 20"},"PeriodicalIF":0.0,"publicationDate":"2023-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"43381776","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}
Compute and memory are tightly coupled within each server in traditional datacenters. Large-scale datacenter operators have identified this coupling as a root cause behind fleetwide resource underutilization and increasing Total Cost of Ownership (TCO). With the advent of ultra-fast networks and cache-coherent interfaces, memory disaggregation has emerged as a potential solution, whereby applications can leverage available memory even outside server boundaries. This paper summarizes the growing research landscape of memory disaggregation from a software perspective and introduces the challenges toward making it practical under current and future hardware trends. We also reflect on our seven-year journey in the SymbioticLab to build a comprehensive disaggregated memory system over ultra-fast networks. We conclude with some open challenges toward building next-generation memory disaggregation systems leveraging emerging cache-coherent interconnects.
{"title":"Memory Disaggregation: Advances and Open Challenges","authors":"H. Maruf, Mosharaf Chowdhury","doi":"10.1145/3606557.3606562","DOIUrl":"https://doi.org/10.1145/3606557.3606562","url":null,"abstract":"Compute and memory are tightly coupled within each server in traditional datacenters. Large-scale datacenter operators have identified this coupling as a root cause behind fleetwide resource underutilization and increasing Total Cost of Ownership (TCO). With the advent of ultra-fast networks and cache-coherent interfaces, memory disaggregation has emerged as a potential solution, whereby applications can leverage available memory even outside server boundaries. This paper summarizes the growing research landscape of memory disaggregation from a software perspective and introduces the challenges toward making it practical under current and future hardware trends. We also reflect on our seven-year journey in the SymbioticLab to build a comprehensive disaggregated memory system over ultra-fast networks. We conclude with some open challenges toward building next-generation memory disaggregation systems leveraging emerging cache-coherent interconnects.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"57 1","pages":"29 - 37"},"PeriodicalIF":0.0,"publicationDate":"2023-05-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"46064963","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}
Application telemetry refers to measurements taken from software systems to assess their performance, availability, correctness, efficiency, and other aspects useful to operators, as well as to troubleshoot them when they behave abnormally. Many modern observability platforms support dimensional models of telemetry signals where the measurements are accompanied by additional dimensions used to identify either the resources described by the telemetry or the business-specific attributes of the activities (e.g., a customer identifier). However, most of these platforms lack any semantic understanding of the data, by not capturing any metadata about telemetry, from simple aspects such as units of measure or data types (treating all dimensions as strings) to more complex concepts such as purpose policies. This limits the ability of the platforms to provide a rich user experience, especially when dealing with different telemetry assets, for example, linking an anomaly in a time series with the corresponding subset of logs or traces, which requires semantic understanding of the dimensions in the respective data sets.
{"title":"Positional Paper","authors":"Y. Shkuro, B. Renard, Ashutosh Kumar Singh","doi":"10.1145/3544497.3544500","DOIUrl":"https://doi.org/10.1145/3544497.3544500","url":null,"abstract":"Application telemetry refers to measurements taken from software systems to assess their performance, availability, correctness, efficiency, and other aspects useful to operators, as well as to troubleshoot them when they behave abnormally. Many modern observability platforms support dimensional models of telemetry signals where the measurements are accompanied by additional dimensions used to identify either the resources described by the telemetry or the business-specific attributes of the activities (e.g., a customer identifier). However, most of these platforms lack any semantic understanding of the data, by not capturing any metadata about telemetry, from simple aspects such as units of measure or data types (treating all dimensions as strings) to more complex concepts such as purpose policies. This limits the ability of the platforms to provide a rich user experience, especially when dealing with different telemetry assets, for example, linking an anomaly in a time series with the corresponding subset of logs or traces, which requires semantic understanding of the dimensions in the respective data sets.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"56 1","pages":"8 - 17"},"PeriodicalIF":0.0,"publicationDate":"2022-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"46420845","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}
Julius Plehn, A. Fuchs, Michael Kuhn, Jakob Lüttgau, T. Ludwig
While compression can provide significant storage and cost savings, its use within HPC applications is often only of secondary concern. This is in part due to the inflexibility of existing approaches where a single compression algorithm has to be used throughout the whole application but also because insights into the behaviour of the algorithms within the context of individual applications are missing.
{"title":"Data-Aware Compression for HPC using Machine Learning","authors":"Julius Plehn, A. Fuchs, Michael Kuhn, Jakob Lüttgau, T. Ludwig","doi":"10.1145/3544497.3544508","DOIUrl":"https://doi.org/10.1145/3544497.3544508","url":null,"abstract":"While compression can provide significant storage and cost savings, its use within HPC applications is often only of secondary concern. This is in part due to the inflexibility of existing approaches where a single compression algorithm has to be used throughout the whole application but also because insights into the behaviour of the algorithms within the context of individual applications are missing.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"56 1","pages":"62 - 69"},"PeriodicalIF":0.0,"publicationDate":"2022-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"46111446","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}
Devashish R. Purandare, Daniel Bittman, E. L. Miller
Modern, large-scale scientific computing runs on complex exascale storage systems that support even more complex data workloads. Understanding the data access and movement patterns is vital for informing the design of future iterations of existing systems and next-generation systems. Yet we are lacking in publicly available traces and tools to help us understand even one system in depth, let alone correlate long-term cross-system trends.
{"title":"Analysis and Workload Characterization of the CERN EOS Storage System","authors":"Devashish R. Purandare, Daniel Bittman, E. L. Miller","doi":"10.1145/3544497.3544507","DOIUrl":"https://doi.org/10.1145/3544497.3544507","url":null,"abstract":"Modern, large-scale scientific computing runs on complex exascale storage systems that support even more complex data workloads. Understanding the data access and movement patterns is vital for informing the design of future iterations of existing systems and next-generation systems. Yet we are lacking in publicly available traces and tools to help us understand even one system in depth, let alone correlate long-term cross-system trends.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"56 1","pages":"55 - 61"},"PeriodicalIF":0.0,"publicationDate":"2022-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"41771130","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}
Yichen Li, Xu Zhang, Shilin He, Zhuangbin Chen, Yu Kang, Jinyang Liu, Liqun Li, Yingnong Dang, Feng Gao, Zhangwei Xu, S. Rajmohan, Qingwei Lin, Dongmei Zhang, Michael R. Lyu
Cloud incidents (service interruptions or performance degradation) dramatically degrade the reliability of large-scale cloud systems, causing customer dissatisfaction and revenue loss. With years of efforts, cloud providers are able to solve most incidents automatically and rapidly. The secret of this ability is intelligent incident detection. Only when incidents are detected timely, accurately, and comprehensively, can they be diagnosed and mitigated at a satisfiable speed. To overcome the limitations of traditional rule-based detection, we carried out years of incident detection research. We developed a comprehensive AIOps (Artificial Intelligence for IT Operations) framework for incident detection containing a set of data-driven methods. This paper shares our recent experience of developing and deploying such an intelligent incident detection system at Microsoft. We first discuss the real-world challenges of incident detection that constitute the pain points of engineers. Then, we summarize our intelligent solutions proposed in recent years to tackle these challenges. Finally, we show the deployment of the incident detection AIOps framework and demonstrate its practical benefits conveyed to Microsoft cloud services with real cases.
{"title":"An Intelligent Framework for Timely, Accurate, and Comprehensive Cloud Incident Detection","authors":"Yichen Li, Xu Zhang, Shilin He, Zhuangbin Chen, Yu Kang, Jinyang Liu, Liqun Li, Yingnong Dang, Feng Gao, Zhangwei Xu, S. Rajmohan, Qingwei Lin, Dongmei Zhang, Michael R. Lyu","doi":"10.1145/3544497.3544499","DOIUrl":"https://doi.org/10.1145/3544497.3544499","url":null,"abstract":"Cloud incidents (service interruptions or performance degradation) dramatically degrade the reliability of large-scale cloud systems, causing customer dissatisfaction and revenue loss. With years of efforts, cloud providers are able to solve most incidents automatically and rapidly. The secret of this ability is intelligent incident detection. Only when incidents are detected timely, accurately, and comprehensively, can they be diagnosed and mitigated at a satisfiable speed. To overcome the limitations of traditional rule-based detection, we carried out years of incident detection research. We developed a comprehensive AIOps (Artificial Intelligence for IT Operations) framework for incident detection containing a set of data-driven methods. This paper shares our recent experience of developing and deploying such an intelligent incident detection system at Microsoft. We first discuss the real-world challenges of incident detection that constitute the pain points of engineers. Then, we summarize our intelligent solutions proposed in recent years to tackle these challenges. Finally, we show the deployment of the incident detection AIOps framework and demonstrate its practical benefits conveyed to Microsoft cloud services with real cases.","PeriodicalId":38935,"journal":{"name":"Operating Systems Review (ACM)","volume":"56 1","pages":"1 - 7"},"PeriodicalIF":0.0,"publicationDate":"2022-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"45597721","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：