Ding Yuan.pdf

Xin Jin.pdf

rachata.pdf

刘峥.pdf

pclee_sketch.pdf

jialin.pd

沈志荣 - 数据中心高可靠纠删码研究.pdf

重新思考Web 场景下的事务抽象-wangzhaoguo.pdf

李永坤 .pd

• INTERNEURON: A Middleware with Multi-Network Communication Reliability for Infrastructure Vehicle Cooperative Autonomous Driving, Tianze Wu
  

• Towards Developing High Performance RISC-V Processors Using Agile Methodology,
  Yinan Xu
  

• NapFS: A High-Performance NUMA-Aware PM File System, Wenqing Jia
  

• FastStore: A High-Performance RDMA-enabled Distributed Key-Value Store with Persistent Memory, ZiWei Xiong
  

• LightShaper：a high-precision and universal traffic shaping tool, Haokun Wang
  

• MCCBench: A C10M Benchmark Oriented to Interactive Network Services, Hui Song
  

• Exploiting Hybrid Index for RDMA-based Key-Value Stores, Shukai Han
  

• GraFF: A Multi-FPGA System with Memory Semantic Fabric for Scalable Graph Processing, Xu Zhang
  

Topic: Software-Defined Cloud Systems
Speaker: Xin Jin, Peking University

Bio: Xin Jin is an Associate Professor in School of Computer Science at Peking University. He received his BS in computer science and BA in economics from Peking University in 2011, and his MA and PhD in computer science from Princeton University in 2013 and 2016. His research is in computer networks and computer systems, with a focus on software-defined datacenters, programmable networks, and cloud computing. He received USENIX FAST Best Paper Award (2019), USENIX NSDI Best Paper Award (2018), Amazon AWS Machine Learning Research Award (2019), Google Faculty Research Award (2019), and Facebook Communications & Networking Research Award (2018).

Abstract: With the end of Moore's law and Dennard's scaling, it is challenging for cloud systems to keep up with the rapidly increasing user demands and meet the Service Level Objectives (SLOs). In this talk, I will present a new paradigm that exploits the capability of domain-specific hardware and co-design network, storage and compute to build next-generation cloud systems and meet these requirements. I will focus on two types of domain-specific hardware, which are programmable switching ASICs for IO-intensive workloads and GPUs for compute-intensive workloads. I will present several systems we have built recently with programmable switching ASICs and GPUs. These systems exemplify a new generation of cloud systems enabled by domain-specific hardware.

Topic: High-performance Memory Management Design for Modern Architectures
Speaker: Rachata Ausavarungnirun, King Mongkut’s University of Technology North Bangkok

Bio: Rachata Ausavarungnirun is an assistant professor at the Sirindhorn International Thai-German Graduate School of Engineering (TGGS) at King Mongkut’s University of Technology North Bangkok. His research spans multiple topics across computer architecture and system software with emphasis on GPU architecture, heterogeneous CPU-GPU architecture, virtual memory, containerization, management of GPUs in the cloud, memory subsystems, memory management, processing-in-memory, non-volatile memory, network-on-chip, and accelerator designs. Rachata received his Ph.D. in Electrical and Computer Engineering from Carnegie Mellon University in 2017.

Abstract: The growth in the new system and architecture designs have enabled significant performance improvement across various types of modern applications. However, these applications' increasing resource demand also creates new challenges as conventional methods to manage virtual memory fail to deliver good performance without non-trivial workarounds across diverse types of architectures. This talk identifies performance bottlenecks created by virtual memory and its metadata management system in modern architectures. Specifically, we provide an in-depth analysis of the bottlenecks and limitations of Linux's huge page on modern applications across multiple architectures. To minimize such performance bottlenecks, we introduce a combination of new techniques with modest hardware changes to manage virtual memory and its metadata. Our proposal allows the system to utilize different policies based on the applications' own demand to eliminate performance pathologies and improves system performance across various architectures.

Topic: 云计算数据中心当下问题与基础软件关键技术
Speaker: 刘峥, 阿里达摩院操作系统实验室

Bio: 刘峥，阿里达摩院操作系统实验室资深技术专家，CCF系统软件专委会委员，龙蜥社区技术委员会委员。2011年加入阿里集团，先后负责操作系统、云计算基础设施、云原生底层系统相关产品的研发工作。当前主要负责下一代数据中心系统软件、软硬件协同优化领域的前沿技术探索等工作。.

Abstract: 云计算的快速普及不断推动着信息产业相关技术的快速发展，从异构算力的支持到全新计算模型的引入，都对数据中心软硬件技术提出了新的挑战。本报告站在基础软件的视角来探讨当下云计算数据中心面临的主要问题，结合阿里在云计算领域的实践，介绍基础软件领域要探索的关键

Topic: Invertible Sketches for Network Measurement at Scale.
Speaker: Patrick P. C. Lee, The Chinese University of Hong Kong

Bio: Patrick P. C. Lee is now a Professor of the Department of Computer Science and Engineering at the Chinese University of Hong Kong. His research interests are in various applied/systems topics on improving the dependability of large-scale storage systems, distributed systems, and networks. He heads the Applied Distributed Systems Lab and is working very closely with a group of graduate students on different projects in networks and systems. He is now serving on the editorial boards of IEEE/ACM Transactions on Networking and ACM Transactions on Storage. For details, please refer to his personal homepage: http://www.cse.cuhk.edu.hk/~pclee.

Abstract: Network measurement in massive network traffic is challenging due to the stringent requirements of fast packet processing and limited resource availability. Given the fast-speed and ever-increasing network traffic, particularly in modern large-scale data center networks, maintaining per-flow state in network measurement tasks inevitably has tremendous resource demands. In this talk, I will introduce invertible sketches for various network measurement tasks that can be carried out in a real-time and scalable manner. Invertible sketches are summary data structures that significantly mitigate memory footprints with bounded errors; by "invertible", we mean that the measurement results can be readily recovered from only the sketch data structure itself. We show how to design invertible sketches for heavy flow detection and superspreader detection.

Topic: Co-Designing Distributed Systems with Programmable Network Hardware
Speaker: Jialin Li, National University of Singapore

Bio: Jialin Li is an Assistant Professor in the School of Computing at the National University of Singapore. He finished his PhD from the University of Washington in 2019, advised by Dan R. K. Ports. As part of his dissertation work, he pioneered a new approach to building high performance distributed systems, by co-designing with data center networks. He is the recipient of best paper awards at OSDI and NSDI. His current research interests include distributed systems and programmable hardware co-design, data plane operating systems, and decentralized system infrastructure.

Abstract:With the end of Dennard scaling and Moore's Law, performance improvement of general purpose processors has been stagnant for the past decade. This is in contrast to the continuous growth in network speed in data centers and telecommunication networks, and the increasing demand of modern applications. Not surprisingly, software processing on CPUs has become the performance bottleneck of many large scale distributed systems deployed in data centers. In this talk, I will introduce a new approach to designing distributed systems in data centers that tackle the aforementioned challenge -- by co-designing distributed systems with the data center network. Specifically, my work has taken advantage of new-generation programmable switches in data centers to build novel network-level primitives with near-zero processing overhead. We then leverage these primitives to enable more efficient protocol and system designs. I will describe several systems we built that demonstrate the benefit of this approach. The first two, Network-Ordered Paxos and Eris, virtually eliminate the coordination overhead in state machine replication and fault-tolerant distributed transactions, by relying on network sequencing primitives to consistently order user requests. The third system, Hydra, overcomes the limitations of a centralized network sequencer in previous solutions, by deploying a distributed set of sequencers without sacrificing the ordering guarantees. The last system, Pegasus, substantially improves the load balancing of a distributed storage system -- up to a 9x throughput improvement over existing solutions -- by implementing an in-network coherence directory in the switching ASICs.

Topic: 数据中心高可靠纠删码技术
Speaker: 沈志荣, 厦门大学

Bio: 沈志荣，厦门大学信息学院副教授，入选厦门大学“南强青年拔尖人才计划”，主要研究方向为大数据/云存储系统、数据中心和新型非易失存储介质的数据存储可靠性，在IEEE TC、TPDS、TDSC和USENIX ATC、INFOCOM、ICDCS、IPDPS、DSN、SRDS等CCF A/B类期刊会议发表论文40多篇，获得SRDS'20最佳论文奖（通讯作者）和SRDS'15最佳论文提名奖（第一作者），主持包括国家重点研发计划课题、国家自然科学基金面上项目和青年基金、CCF-腾讯犀牛鸟基金和CCF-华为胡杨林基金等多项课题。

Abstract: 数据中心软硬件故障成为常态，数据存储的可靠性成为数据中心的一个重要问题。纠删码是一种具有低存储开销的容错编码，如今被广泛应用于Hadoop、Windows Azure Storage、Ceph等开源或商用存储系统之中。然而纠删码具有修复代价高和更新开销大等问题，从而极易激发大量的网络和存储I/O。在该问题上，我们展开了一些初步探索，分别基于数据中心网络架构和纠删码理论，设计了一系列降低数据中心纠删码修复和校验更新开销等方法，从而能够显著加速数据中心的数据修复和校验更新操作。

Topic: 重新思考Web 场景下的事务抽象与SQL优化问题
Speaker: 王肇国, 上海交通大学

Bio:  上海交通大学长聘教轨副教授，博士生导师。于2008年从南京大学本科毕业，11年和14年分别从复旦大学获得硕士和博士学位。2014至2018年在纽约大学从事博士后研究工作，后担任研究助理教授。主要从事并行与分布式数据库系统方面的研究，主攻系统一致性、存储结构和SQL优化方面问题。相关成果发表在SIGMOD，OSDI， PPoPP，EuroSys, Usenix ATC, NSDI，PODC等国际重要会议上。曾获SIGMOD 2022最佳论文优胜奖（Honorable Mention），APSys 2017最佳论文奖，ACM ChinaSys新星奖，华为奥林帕斯先锋奖，以及两次华为火花奖。他是CCF数据库专委执行委员、ACM ChinaSys青执委委员、开源组委委员、FCS预备青年编委、ACM SIGOPS指定编辑、ChinaSys学术开源创新平台的主要发起人，曾受邀担任 ACM ChinaSys 2021大会联合主席、ACM ChinaSys 2022程序委员会联合主席，ACM APSys 2018宣传主席， IEEE ICDCS 2020、IEEE Cluster 2021、TPDS 专刊 2020/2021等国际会议和专刊的程序委员会成员。

Abstract:如何让数据库系统更好的服务当代Web应用？这一直是学术界与工业界共同关心的问题。针对该问题，我们从事务抽象和SQL优化两方面切入，对Web场景下传统数据库技术的合理性进行了大胆探究。在事务抽象方面，我们分析了GitHub上多个大型Web应用，系统性的揭示了Web应用中的即席事务（Ad Hoc Transaction）现象以及其对应用和系统的影响；在SQL优化方面，我们借鉴超优化思想，根据Web场景需求，自动生成SQL重写规则（Rewrite Rule），有效提高了对Web应用中SQL优化效率。

Topic: 高性能高可靠键值存储系统
Speaker: 李永坤, 中国科学技术大学

Bio: 李永坤，博士，中国科学技术大学副教授，中科院青促会会员，仲英青年学者。主要研究方向是存储系统，包括键值存储、内存系统、虚拟化与图计算等。目前共发表论文60余篇。主持国家自然科学基金青年与面上各一项、科技部重点研发计划子课题一项，以及PingCAP、华为等多项企业合作项目。部分成果在PingCAP落地应用，获评PingCAP优秀合作奖。担任ASPLOS2022(ERC)、APSys2020等会议程序委员会委员，以及Journal of Computer Science and Technology青年编委。

Abstract: 键值存储可以有效支持非结构化数据存储，并提供良好的横向扩展性。持久化键值存储系统广泛采用基于LSM-tree的存储结构进行数据管理，通过存储分层、日志写、数据排序等思想，实现高效的数据写入，以及优良的单点查询与范围查询性能。但是，基于LSM-tree的键值存储由于需要维护数据有序并需要多层查找，面临严重的读写放大问题，尤其是在分布式存储场景，数据往往需要存储多份以保证可靠性，从而严重加剧基于LSM-tree的键值存储的读写放大问题。本报告主要介绍我们对基于多副本容错的键值存储优化，尤其是多副本解耦的存储结构设计等，以实现高性能高可靠的键值存储系统。