基于FPGA的100Gbps以太网键值存储高吞吐量低延迟分布式管理代理

Jinyu Xie, Yunhui Qiu, Wenbo Yin, Lingli Wang
{"title":"基于FPGA的100Gbps以太网键值存储高吞吐量低延迟分布式管理代理","authors":"Jinyu Xie, Yunhui Qiu, Wenbo Yin, Lingli Wang","doi":"10.1109/ICFPT47387.2019.00034","DOIUrl":null,"url":null,"abstract":"Key-value store (KVS) is one of the most important NoSQL databases. To satisfy the demand for growing data in the era of big data, distributed KVS systems that enhance the scalability of storage systems are widely used in recent years. Hence, high-performance distributed management that implements functions such as data partitioning and message distribution is indispensable. In this paper, we present a high-throughput and low-latency distributed management proxy (DMP) for key-value store over 100Gbps Ethernet (100GbE). The proxy-based design which has flexibility and generality is fully implemented on FPGA. In the DMP, the consistent hashing algorithm is designed to partition KVS packets in pipeline. Then a hardware network offload engine is proposed to transmit network messages over 100GbE on FPGA to reduce the latency penalties caused by consistent hashing. Besides, in order to utilize the 100Gbps Ethernet bandwidth, four fully parallel and pipelined data paths are implemented in the DMP on an FPGA. The experimental results show that the peak throughput of the proposed DMP can reach 77.4 million queries per second (QPS), and the latency can achieve as low as 0.8µs for the small-size packets, which is 7.4x higher and 23x faster respectively than the software proxy implementation accelerated by Intel's DPDK.","PeriodicalId":241340,"journal":{"name":"2019 International Conference on Field-Programmable Technology (ICFPT)","volume":"520 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":"{\"title\":\"High-Throughput and Low-Latency Distributed Management Proxy for Key-Value Store Over 100Gbps Ethernet on FPGA\",\"authors\":\"Jinyu Xie, Yunhui Qiu, Wenbo Yin, Lingli Wang\",\"doi\":\"10.1109/ICFPT47387.2019.00034\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Key-value store (KVS) is one of the most important NoSQL databases. To satisfy the demand for growing data in the era of big data, distributed KVS systems that enhance the scalability of storage systems are widely used in recent years. Hence, high-performance distributed management that implements functions such as data partitioning and message distribution is indispensable. In this paper, we present a high-throughput and low-latency distributed management proxy (DMP) for key-value store over 100Gbps Ethernet (100GbE). The proxy-based design which has flexibility and generality is fully implemented on FPGA. In the DMP, the consistent hashing algorithm is designed to partition KVS packets in pipeline. Then a hardware network offload engine is proposed to transmit network messages over 100GbE on FPGA to reduce the latency penalties caused by consistent hashing. Besides, in order to utilize the 100Gbps Ethernet bandwidth, four fully parallel and pipelined data paths are implemented in the DMP on an FPGA. The experimental results show that the peak throughput of the proposed DMP can reach 77.4 million queries per second (QPS), and the latency can achieve as low as 0.8µs for the small-size packets, which is 7.4x higher and 23x faster respectively than the software proxy implementation accelerated by Intel's DPDK.\",\"PeriodicalId\":241340,\"journal\":{\"name\":\"2019 International Conference on Field-Programmable Technology (ICFPT)\",\"volume\":\"520 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2019-12-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"1\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2019 International Conference on Field-Programmable Technology (ICFPT)\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1109/ICFPT47387.2019.00034\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2019 International Conference on Field-Programmable Technology (ICFPT)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ICFPT47387.2019.00034","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1

摘要

键值存储(KVS)是最重要的NoSQL数据库之一。为了满足大数据时代对数据增长的需求,分布式KVS系统增强了存储系统的可扩展性,近年来得到了广泛的应用。因此,实现数据分区和消息分发等功能的高性能分布式管理是必不可少的。在本文中,我们提出了一种用于100Gbps以太网(100GbE)上键值存储的高吞吐量和低延迟分布式管理代理(DMP)。在FPGA上充分实现了基于代理的设计,具有灵活性和通用性。在DMP中,采用一致性哈希算法对管道中的KVS包进行分区。然后提出了一种硬件网络卸载引擎,在FPGA上传输100GbE以上的网络消息,以减少一致性哈希带来的延迟损失。此外,为了充分利用100Gbps的以太网带宽,在FPGA上的DMP中实现了四条完全并行和流水线的数据路径。实验结果表明,该DMP的峰值吞吐量可达7740万次查询/秒(QPS),对于小尺寸数据包的延迟可低至0.8µs,分别比由英特尔DPDK加速的软件代理实现高7.4倍和快23倍。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
High-Throughput and Low-Latency Distributed Management Proxy for Key-Value Store Over 100Gbps Ethernet on FPGA
Key-value store (KVS) is one of the most important NoSQL databases. To satisfy the demand for growing data in the era of big data, distributed KVS systems that enhance the scalability of storage systems are widely used in recent years. Hence, high-performance distributed management that implements functions such as data partitioning and message distribution is indispensable. In this paper, we present a high-throughput and low-latency distributed management proxy (DMP) for key-value store over 100Gbps Ethernet (100GbE). The proxy-based design which has flexibility and generality is fully implemented on FPGA. In the DMP, the consistent hashing algorithm is designed to partition KVS packets in pipeline. Then a hardware network offload engine is proposed to transmit network messages over 100GbE on FPGA to reduce the latency penalties caused by consistent hashing. Besides, in order to utilize the 100Gbps Ethernet bandwidth, four fully parallel and pipelined data paths are implemented in the DMP on an FPGA. The experimental results show that the peak throughput of the proposed DMP can reach 77.4 million queries per second (QPS), and the latency can achieve as low as 0.8µs for the small-size packets, which is 7.4x higher and 23x faster respectively than the software proxy implementation accelerated by Intel's DPDK.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信