https://sazikk.top/A minimal, responsive and feature-rich Jekyll theme for technical writing. 2026-01-11T23:15:18+08:00 sazikk https://sazikk.top/ Jekyll © 2026 sazikk /assets/img/favicons/favicon.ico /assets/img/favicons/favicon-96x96.png 2026-01-07T00:00:00+08:00 2026-01-07T00:00:00+08:00 https://sazikk.top/posts/%E7%AC%94%E8%AE%B0-%E8%AE%BA%E6%96%87%E9%98%85%E8%AF%BB%E7%AC%94%E8%AE%B0-Scalable-Far-Memory-Balancing-Faults-and-Evictions/ sazikk

Scalable Far Memory: Balancing Faults and Evictions Yueyang Pan (EPFL), Yash Lala (Yale University), Musa Unal (EPFL), Yujie Ren (EPFL), Seung-seob Lee (Yale University), Abhishek Bhattacharjee (Yale University), Anurag Khandelwal (Yale University), Sanidhya Kashyap (EPFL) 背景 基于分页的 far memory 系统可以让应用程序透明的访存，但是目前的 SOTA far memory 系统（Hermit、DiLOS）无法应对较大规模的情况，这会使他们不适用于一些既需要远程内存支持规模又随线程数量增长的计...

2025-12-15T00:00:00+08:00 2025-12-15T00:00:00+08:00 https://sazikk.top/posts/%E7%AC%94%E8%AE%B0-%E8%AE%BA%E6%96%87%E9%98%85%E8%AF%BB%E7%AC%94%E8%AE%B0-DRack-A-CXL-Disaggregated-Rack-Architecture-to-Boost-Inter-Rack-Communication/ sazikk

DRack：A CXL-Disaggregated Rack Architecture to Boost Inter-Rack Communication Xu Zhang (ICT, CAS & University of Chinese Academy of Sciences), Ke Liu (ICT, CAS & University of Chinese Academy of Sciences), Yuan Hui (Huawei), Xiaolong Zheng (Huawei), Yisong Chang (ICT, CAS & University of Chinese Academy of Sciences), Yizhou Shan (Huawei Cloud), Guanghui Zhang (Shandong Universit...

2025-12-05T00:00:00+08:00 2025-12-05T00:00:00+08:00 https://sazikk.top/posts/%E7%AC%94%E8%AE%B0-%E8%AE%BA%E6%96%87%E9%98%85%E8%AF%BB%E7%AC%94%E8%AE%B0-Crux-GPU-Efficient-Communication-Scheduling-for-Deep-Learning-Training/ sazikk

Crux：GPU-Efficient Communication Scheduling for Deep Learning Training Jiamin Cao (Alibaba Cloud), Yu Guan (Alibaba Cloud), Kun Qian (Alibaba Cloud), Jiaqi Gao (Alibaba Cloud), Wencong Xiao (Alibaba Cloud), Jianbo Dong (Alibaba Cloud), Binzhang Fu (Alibaba Cloud), Dennis Cai (Alibaba Cloud), Ennan Zhai (Alibaba Cloud) 背景 DLT(Deep Learning Traininng) 任务已经成为了多租户云计算服务中最重要的服务。作者团队发现，在同一集群上运行多...

2025-11-26T00:00:00+08:00 2026-01-04T22:03:25+08:00 https://sazikk.top/posts/%E7%AC%94%E8%AE%B0-%E8%AE%BA%E6%96%87%E9%98%85%E8%AF%BB%E7%AC%94%E8%AE%B0-Spirit-Fair-Allocation-of-Interdependent-Resources-in-Remote-Memory-Systems/ sazikk

Spirit: Fair Allocation of Interdependent Resources in Remote Memory Systems Seung-seob Lee(Yale University), Jachym Putta (Yale University), Ziming Mao (UC Berkeley), Anurag Khandelwal (Yale University) 背景 多用户 RDMA 系统中的资源分配一直是问题，对于同一台机器上的多个 RDMA 用户，他们各自分配有一部分带宽和 dram 作为访问 far memory 的 cache。但是，大 cache 意味着应用可能不需要频繁的和 far memory 通信，仅需要较小的带宽；而大带宽意味着应用可以以更高的频率和 far memory 进行通信，cache 可以较小。 ...

2025-09-17T00:00:00+08:00 2025-09-18T20:40:56+08:00 https://sazikk.top/posts/%E7%AC%94%E8%AE%B0-%E8%AE%BA%E6%96%87%E9%98%85%E8%AF%BB%E7%AC%94%E8%AE%B0-Low-Overhead-Distributed-Application-Observation-with-DeepTrace-Achieving-Accurate-Tracing-in-Production-Systems/ sazikk

Low-Overhead Distributed Application Observation with DeepTrace: Achieving Accurate Tracing in Production Systems Yantao Geng (Tsinghua University), Han Zhang (Tsinghua University), Zhiheng Wu (Tsinghua University), Yahui Li (Tsinghua University), Jilong Wang (Tsinghua University), Xia Yin (Tsinghua University) 背景 微服务的发展使得其调试越来越困难，需要一个完善的分布式 trace 框架。现有的非侵入式 trace 框架主要采取两种方案：基于 FIFO 的关联框架...