Microsoft at EuroSys 2023: Systems innovation across the stack to help support an easier, faster, safer, and smarter cloud

EuroSys 2023 (opens in new tab) is the premier systems conference in Europe, and 2023 marks its 18th edition. Sponsored by ACM SIGOPS Europe and hosted May 8 to May 12, the conference covers a wide range of topics, including operating systems, real-time and networked systems, storage and middleware, and distributed, parallel, and embedded computing, as well as their implications for applications and hardware.

As in previous years, Microsoft has a strong presence in the conference, drawing from research and production teams in Asia, Europe, and the United States, including Azure Systems Research, in collaboration with many universities. This work spans areas including systems for machine learning, serverless computing, datacenter networking, caching, and debugging. We’re also participating in several of the associated workshops and in key aspects of the organization, including Senior Principal Researcher Dushyanth Narayanan as the program co-chair for the main conference.

Here are some of the highlights (see below for more information about Microsoft at EuroSys, including the authors of the published papers):

Datacenter networking

The paper «Saba: Rethinking Datacenter Network Allocation from Application’s Perspective» proposes allocating datacenter network bandwidth according to applications’ sensitivity to bandwidth, achieving significant performance gains compared to fair sharing. In «FlexPass: A Case for Flexible Credit-based Transport for Datacenter Networks,» the authors make the case to incrementally deploy new proactive, credit-based transport protocols in the datacenter.

Serverless computing

In serverless computing, the paper «Palette Load Balancing: Locality Hints for Serverless Functions (opens in new tab)» proposes adding locality to Function-as-a-Service (FaaS) serverless systems, closing the performance gap between serverful data-intensive applications and their serverless implementation. In «With Great Freedom Comes Great Opportunity: Rethinking Resource Allocation for Serverless Functions (opens in new tab),» the authors revisit the FaaS interface and find that correctly choosing memory, CPU, and architecture for each serverless function can allow both providers and customers to improve cost and performance. Finally, in «Groundhog: Efficient Request Isolation in FaaS (opens in new tab),» the authors present a system that enables efficient snapshots for better isolation between function invocations. Microsoft is also represented in the SErverless Systems, Applications and MEthodologies (SESAME) workshop (opens in new tab) with a keynote and work-in-progress paper.

Concurrency debugging

In «WAFFLE: Exposing Memory Ordering Bugs Efficiently with Active Delay Injection,» the authors tackle the difficult problem of finding memory ordering bugs, a type of concurrency bug caused by incorrect timing between a memory access to a particular object and the object’s initialization or deallocation. Their proposed tool uses delay injection techniques and, through key innovations, can expose more bugs with less overhead than the state of the art.

Systems for machine learning

In «SiloD: A Co-design of Caching and Scheduling for Deep Learning Clusters,» the proposed framework treats cache and remote I/O as first-class resources and can integrate different state-of-the-art deep learning scheduling policies in a unified scheduling framework. 

Caching

In «FrozenHot Cache: Rethinking Cache Management for Modern Hardware,» the authors introduce a generic approach to improve the scalability of traditional list-based caches, such as least recently used (LRU), by separating the objects into two regions: a frozen region that serves requests for hot objects with minimal latency by eliminating promotion and locking and a regular dynamic region that uses the existing cache design to achieve workload adaptivity. 

Microsoft papers published at EuroSys with their authors:

1. Saba: Rethinking Datacenter Network Allocation from Application’s Perspective
   M.R. Siavash Katebzadeh, University of Edinburgh; Paolo Costa, Microsoft Research; Boris Grot, University of Edinburgh
2. FlexPass: A Case for Flexible Credit-based Transport for Datacenter Networks
   Hwijoon Lim, Jaehong Kim, KAIST; Inho Cho, MIT CSAIL; Keon Jang, MPI-SWS, Rubrik; Wei Bai, Microsoft Research; Dongsu Han, KAIST
3. Palette Load Balancing: Locality Hints for Serverless Functions
   Mania Abdi, Northeastern University; Samuel Ginzburg, Princeton; Xiayue Charles Lin, Anyscale; Jose Faleiro, unaffiliated; Gohar Irfan Chaudhry, Íñigo Goiri, Ricardo Bianchini, Daniel S. Berger, Rodrigo Fonseca, Azure Systems Research
4. With Great Freedom Comes Great Opportunity: Rethinking Resource Allocation for Serverless Functions
   Muhammad Bilal, Instituto Superior Técnico (ULisboa), INESC-ID, UCLouvain; Marco Canini, KAUST; Rodrigo Fonseca, Azure Systems Research; Rodrigo Rodrigues, Instituto Superior Técnico (ULisboa), INESC-ID
5. Groundhog: Efficient Request Isolation in FaaS
   Mohamed Alzayat, Max Planck Institute for Software Systems (MPI-SWS); Jonathan Mace, Microsoft Research; Peter Druschel, Deepak Garg, Max Planck Institute for Software Systems (MPI-SWS)
6. WAFFLE: Exposing Memory Ordering Bugs Efficiently with Active Delay Injection
   Bogdan Alexandru Stoica, Shan Lu, University of Chicago; Madanlal Musuvathi, Suman Nath, Microsoft Research
7. SiloD: A Co-design of Caching and Scheduling for Deep Learning Clusters
   Hanyu Zhao, Peking University; Zhenhua Han, Microsoft Research; Zhi Yang, Peking University; Quanlu Zhang, Microsoft Research; Mingxia Li, USTC; Fan Yang, Qianxi Zhang, Microsoft Research; Binyang Li, Microsoft; Yuqing Yang, Lili Qiu, Microsoft Research; Lintao Zhang, BaseBit Technologies; Lidong Zhou, Microsoft Research
8. FrozenHot Cache: Rethinking Cache Management for Modern Hardware
   Ziyue Qiu, University of Science and Technology of China, Microsoft Research, Carnegie Mellon University; Juncheng Yang, Carnegie Mellon University; Juncheng Zhang, University of Science and Technology of China; Cheng Li, University of Science and Technology of China, Anhui Province Key Laboratory of High Performance Computing; Xiaosong Ma, Qatar Computing Research Institute, HBKU; Qi Chen, Mao Yang, Microsoft Research; Yinlong Xu, University of Science and Technology of China, Anhui Province Key Laboratory of High Performance Computing

EuroSys 2023 Organization Committee:

• Dushyanth Narayanan, Co-Chair
• Jayashree Mohan, Publicity Chair

Program Committee:

• Landon Cox
• Tim Harris
• Marios Kogias
• Jayashree Mohan
• Fan Yang

SESAME Workshop

• Keynote: Rodrigo Fonseca, Azure Systems Research
• Work in Progress: The Neglected Cost of Serverless Cluster Management
  Lazar Cvetković, ETH Zürich; Rodrigo Fonseca, Azure Systems Research; Ana Klimovic, ETH Zürich

EuroSys Doctoral Workshop

• Program Committee: Rodrigo Fonseca, Azure Systems Research

PaPoC Workshop

• Trees and Turtles: Modular Abstractions for State Machine Replication Protocols
  Natalie Neamtu, Microsoft; Haobin Ni, Robbert van Renesse, Cornell University

EuroMLSys Workshop

• Program Committee: Fiodar Kazhamiaka, Azure Systems Research, Stanford University

SysTEX Workshop

• Program Committee: Stavros Volos, Microsoft Research

S’ouvre dans un nouvel onglet