José L. Abellán, Alberto Ros, Juan Fernández Peinador, M. Acacio
{"title":"在多核cmp中表达一致性通知以实现高效的缓存一致性","authors":"José L. Abellán, Alberto Ros, Juan Fernández Peinador, M. Acacio","doi":"10.1109/SAMOS.2013.6621128","DOIUrl":null,"url":null,"abstract":"It is commonly stated that a directory-based coherence protocol is the design of choice to provide maximum performance in coherence maintenance for shared-memory many-core CMPs. Nevertheless, new solutions are emerging to achieve acceptable levels of on-chip area overhead and energy consumption to also meet scalability. In this work, we propose the Express COherence NOtification (ECONO) protocol, a coherence protocol aimed at providing high performance with minimal on-chip area and energy consumption for superior scalability. To maintain coherence, ECONO relies on express coherence notifications which are broadcast atomically over a dedicated lightweight and power-efficient on-chip network leveraging state-of-the-art technology. We implement and evaluate ECONO utilizing full-system simulation, a representative set of benchmarks, and compare it against two contemporary coherence protocols: Hammer and Directory. While ECONO achieves slightly better performance than Directory, our proposal does not need to encode sharer sets like in Hammer, saving significant on-chip area and energy even when considering the extra hardware resources required by ECONO. Projections for a 1024-core CMP reveal that, in comparison to one of the most scalable directory-based protocols to date, ECONO entails more than 2× less on-chip storage overhead while keeping with reasonable power dissipation.","PeriodicalId":382307,"journal":{"name":"2013 International Conference on Embedded Computer Systems: Architectures, Modeling, and Simulation (SAMOS)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2013-07-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"4","resultStr":"{\"title\":\"ECONO: Express coherence notifications for efficient cache coherency in many-core CMPs\",\"authors\":\"José L. Abellán, Alberto Ros, Juan Fernández Peinador, M. Acacio\",\"doi\":\"10.1109/SAMOS.2013.6621128\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"It is commonly stated that a directory-based coherence protocol is the design of choice to provide maximum performance in coherence maintenance for shared-memory many-core CMPs. Nevertheless, new solutions are emerging to achieve acceptable levels of on-chip area overhead and energy consumption to also meet scalability. In this work, we propose the Express COherence NOtification (ECONO) protocol, a coherence protocol aimed at providing high performance with minimal on-chip area and energy consumption for superior scalability. To maintain coherence, ECONO relies on express coherence notifications which are broadcast atomically over a dedicated lightweight and power-efficient on-chip network leveraging state-of-the-art technology. We implement and evaluate ECONO utilizing full-system simulation, a representative set of benchmarks, and compare it against two contemporary coherence protocols: Hammer and Directory. While ECONO achieves slightly better performance than Directory, our proposal does not need to encode sharer sets like in Hammer, saving significant on-chip area and energy even when considering the extra hardware resources required by ECONO. Projections for a 1024-core CMP reveal that, in comparison to one of the most scalable directory-based protocols to date, ECONO entails more than 2× less on-chip storage overhead while keeping with reasonable power dissipation.\",\"PeriodicalId\":382307,\"journal\":{\"name\":\"2013 International Conference on Embedded Computer Systems: Architectures, Modeling, and Simulation (SAMOS)\",\"volume\":\"17 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2013-07-15\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"4\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2013 International Conference on Embedded Computer Systems: Architectures, Modeling, and Simulation (SAMOS)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/SAMOS.2013.6621128\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2013 International Conference on Embedded Computer Systems: Architectures, Modeling, and Simulation (SAMOS)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/SAMOS.2013.6621128","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
ECONO: Express coherence notifications for efficient cache coherency in many-core CMPs
It is commonly stated that a directory-based coherence protocol is the design of choice to provide maximum performance in coherence maintenance for shared-memory many-core CMPs. Nevertheless, new solutions are emerging to achieve acceptable levels of on-chip area overhead and energy consumption to also meet scalability. In this work, we propose the Express COherence NOtification (ECONO) protocol, a coherence protocol aimed at providing high performance with minimal on-chip area and energy consumption for superior scalability. To maintain coherence, ECONO relies on express coherence notifications which are broadcast atomically over a dedicated lightweight and power-efficient on-chip network leveraging state-of-the-art technology. We implement and evaluate ECONO utilizing full-system simulation, a representative set of benchmarks, and compare it against two contemporary coherence protocols: Hammer and Directory. While ECONO achieves slightly better performance than Directory, our proposal does not need to encode sharer sets like in Hammer, saving significant on-chip area and energy even when considering the extra hardware resources required by ECONO. Projections for a 1024-core CMP reveal that, in comparison to one of the most scalable directory-based protocols to date, ECONO entails more than 2× less on-chip storage overhead while keeping with reasonable power dissipation.