SPAIL
System Performance Analytics & Intelligence Lab
浙江大学软件学院
Data-Driven Performance · Hardware-Aware Intelligence
关于我们
SPAIL是浙江大学软件学院的研究实验室，专注于解决云、AI和大数据领域的真实性能瓶颈。实验室由具备顶尖学术背景和丰富产业经验的资深工程师团队领导，与多家科技巨头保持深度产学研合作。
核心优势：
- 产业级导师团队：导师均来自Intel、阿里巴巴、腾讯等头部企业，拥有双11、腾讯会议等大规模系统优化实战经验
- 真实场景研究：合作项目直接来自华为、字节跳动、快手等行业伙伴
- 充足资源保障：充足的科研经费支持，提供实习、交流和职业发展机遇
团队
实验室主任
周经森 (Kingsum Chow)
浙江省顶尖人才，前阿里巴巴首席科学家、Intel高级首席工程师。Java全球标准委员会JCP-EC首位中国成员，拥有中美专利72+项，发表论文135+篇。
GitHub 主页 | 浙江大学个人主页
核心研究员
- 吴克强：前Intel/Oracle，芯片缓存设计与AI性能优化专家，10+项美国专利
- 赵海亮：浙大”百人计划”研究员，服务计算与调度优化方向，主持国自然青年项目
- 常志豪：浙大特聘研究员，软硬件协同优化专家，前阿里巴巴技术专家
- 张金山：浙大特聘研究员，CCF专业组委员，IJCAI/EC等顶会论文40+篇
- 智晨：浙大特聘副研究员，智能化软件工程方向，主持国自然青年基金
- 温利龙：浙大计算机博士，系统性能优化与多模态大模型方向
- 黄益聪：高级工程师，前Intel/阿里巴巴/美柚科技技术VP，15项中国专利+3项美国专利
- 李成栋：高级工程师，前腾讯T11/阿里巴巴专家，主导腾讯会议、TDSQL等核心产品性能优化
研究领域
- PIPA 统一性能分析平台 GitHub仓库
- 可观测工具：多源性能数据采集（perf/sar/eBPF等）、跨层深度观测（系统/代码/指令级）、数据质控标准化（验证/清洗/标签）、可视化报告输出
- 软硬件协同优化：跨架构适配（x86_64/ARM/RISC-V）、全栈跨层瓶颈归因、软硬件资源与执行联动分析、针对性优化建议
- 软硬件协同优化
- 微架构分析与性能/功耗优化
- 全栈软件优化（应用/中间件/编译器）
- Java应用与JVM动态调优
- AI系统优化
- 基于大模型的自动优化
- 深度学习模型瓶颈分析与加速
- 基准测试与表征
- 行业基准性能提升
- 工作负载特征分析
- RISC-V/Arm生态性能优化
合作企业
阿里巴巴、华为、腾讯、字节跳动、快手、Ampere Computing、Intel、Oracle、Microsoft
代表性项目：
- 阿里巴巴集群CPU利用率优化与修正算法
- 华为鸿蒙系统性能优化
- 腾讯会议大规模并发性能提升（单房间3倍扩容）
- 字节跳动/快手分布式系统性能分析
开源资源
- PIPA: GitHub仓库
加入我们
招生方向：硕士/博士研究生、研究助理、博士后
要求：计算机、软件工程、电子信息等相关专业，对系统优化有浓厚兴趣
优势：接触真实工业数据、参与顶级会议论文、推荐大厂实习/就业机会
申请方式：发送简历至 ksumchow@outlook.com
联系方式
- 地址：浙江省宁波市鄞州区浙江大学软件学院
- 邮箱：ksumchow@outlook.com
- GitHub: github.com/ZJU-SPAIL
KinsumChow
Kingsum Chow (周经森)
Researcher, Lab Director, & Enterprise CEO
"Software Hardware Co-optimization & System Performance Analytics"
1️⃣ Biography & Impact
🎓 Current Role & Education
- Researcher / Ph.D. Supervisor – School of Software Technology, Zhejiang University
- Director – SPAIL (System Performance Analytics and Intelligence Lab)
- Ph.D. – Computer Science & Engineering, University of Washington, 1996
Advisor: ACM/IEEE Fellow David Notkin
💼 Industry Career
- Chief Scientist – Alibaba (2016–2022)
- Principal Engineer – Intel Corporation, USA (1996–2016)
🚀 Technical & Economic Impact
- Focus: Software-Hardware Co-optimization (SHCO), Performance Analytics & Intelligence
- Accumulated industry savings: > 💰USD 20 billion
- Scale: Optimized tens of millions of servers worldwide, including Double-11 peak workloads
🌐 Global Authority
- Java Standards: First and only Chinese member, JCP-EC (2018–2022)
- Publications: 135+ papers; 74 patents (24 granted US patents)
2️⃣ Research & Subject
🔗SPAIL Lab (System Performance Analytics and Intelligence Lab)
Leading a team of industry veterans and top researchers to solve bottlenecks in Cloud, AI, and Big Data.
🔗PIPA - SPAIL
Platform for Integrated Performance Analytics A unified framework designed to describe, analyze, and optimize system performance across heterogeneous architectures.
3️⃣ Projects & Collabrations
Dr. Chow has led large-scale, high-impact collaborations with global technology leaders, demonstrating expertise in full-stack system optimization. The projects he has spearheaded accumulated an astonishing total budget exceeding 💰160 million CNY (over 💰20 million USD).
- Strategic Ecosystem Partnerships: Collaborated extensively with industry giants including Amazon, Ampere, Arm, Google, Huawei, Microsoft, Tencent, and Meta.
- Project Apollo (Intel & Oracle, 2014–2016): Led the collaboration for the 2015 Oracle Cloud launch, which was announced by the CEOs of both companies.
- Alibaba SPEED (2018–2020): Led the development of the “System Performance Estimation, Evaluation and Decision” platform for Alibaba.
- Project Meta (Intel & Meta, 2022–2023): A major leadership initiative with a vast budget focused on advanced system research.
- Huawei Software Performance Optimization (2024–2026): Leading a multi-year project dedicated to optimizing Huawei’s core software performance.
- Heterogeneous Serverless Optimization (2024–2026): Focused on performance modeling and optimization for serverless, GPU throughput, and microservice environments, collaborating with Alibaba, Kuaishou, ByteDance, and Ampere.
- Alibaba Dragonwell JDK (2018–2019): Spearheaded the development and optimization of Alibaba’s critical Java Development Kit.
- Oracle Exalytics Memory Optimization (2013–2014): Led performance optimization for Oracle’s in-memory analytics system.
- Intel P6 Microcode Simulator (1993–1994): Early high-impact work involving the development of a performance simulator for Intel’s P6 microcode.
4️⃣ Keynote Presentations
I have delivered keynotes at major industry conferences, including 4 appearances at JavaOne, the world’s highest-rated Java conference.
- CMG IMPACT 2022: Propelling Java at Alibaba Scale (Jan 2022)
- QCon Shanghai 2021: Toward Software Performance Evaluation at Scale: A Journey (Link) (Oct 2021)
- Arm DevSummit: Keynote Presentation (Nov 2020 & Oct 2020)
- QCon Beijing: Keynote (2017)
- JavaOne (San Francisco): Keynote Speaker (2017, 2011, 2008, 2007)
-
JavaOne Keynote (2017) 
QCon Shanghai Keynote (2021)
5️⃣ Patents
🇺🇸 Granted US Patents（24）
- US10762065 – Performance monitoring
- US10452443 – Dynamic tuning of a multi-processor/core computing system
- US10120731 – Methods and apparatus to measure hardware performance
- US10102134 – Instructions and logic for run-time evaluation of multiple prefetchers
- US10089207 – Performance variation estimation for applications
- US9954744 – Estimating performance variation of an application without prior knowledge
- US9760404 – Dynamic performance optimization for multi-core systems
- US9639884 – Adaptive prefetch throttling
- US9589024 – Performance-aware resource allocation
- US9378021 – Cache management for virtualized environments
- US9286224 – Throttling prefetch requests for a processor socket
- US9223699 – Method and apparatus for energy-efficient prefetching
- US8583507 – Performance counter virtualization
- US8321290 – Business process and apparatus for online buying using rule-based transferable baskets
- US7542924 – Apparatus for dynamic binary translation
- US7454523 – Method for low-overhead performance monitoring
- US7216154 – Apparatus and method for facilitating access to network resources
- US7032017 – System and method for predictive resource allocation
- US6850899 – Method for high-accuracy branch prediction
- US6772324 – Processor having program counter and execution pipeline external trace buffers
- US6741990 – Trace-driven workload characterization
- US6684252 – Method and system for predicting computer-server performance
- US6493820 – System for online performance diagnostics
- US6182210 – Method and apparatus for real-time performance tuning
🇺🇸 Published US Applications（22）
- US20210056086 – Cross-architecture performance projection
- US20170337083 – Cloud-scale performance regression detection
- US20170169064 – Adaptive sampling for large-scale systems
- US20170060635 – Method for updating software with zero downtime
- US20170063652 – Hardware-assisted performance tracing
- US20160299847 – Energy-aware workload scheduling
- US20150378861 – Performance anomaly detection using ML
- US20150234663 – Cache partitioning for multi-tenant systems
- US20150220372 – Method for fast micro-benchmark synthesis
- US20150220528 – Scalable performance counters
- US20150149714 – Dynamic voltage/frequency control
- US20140281230 – Cross-platform binary instrumentation
- US20140222617 – Hardware-support for managed-runtime profiling
- US20130103541 – Predictive power management
- US20090307108 – Method for scalable event tracing
- US20050131772 – System for automated bottleneck analysis
- US20030097412 – Method for high-resolution time measurement
- US20030061360 – Framework for continuous performance validation
- US20030033511 – Adaptive feedback-driven optimization
- US20020178169 – System for heterogeneous workload co-location
- US20020143991 – Method for lightweight memory profiling
- US20010014941 – Early-stage performance modeling
🇨🇳 中国专利（已公开/授权）
- CN111435317B – 数据处理方法、计算设备及存储介质（发明人：郭健美、周经森；权利人：阿里巴巴集团；已授权）
- CN110998539B – 系统更新的性能影响分析（发明人：周经森、朱婉怡；权利人：阿里巴巴集团；已授权）
- CN110235085A – 确定多处理系统的处理器使用率（发明人：周经森等；权利人：阿里巴巴集团）
- CN110741351A – 确定虚拟化多处理系统的处理器利用率
- CN105164651A – 在管理的运行时间环境域中的高速缓存管理（权利人：英特尔）
- CN111435317A – 数据处理方法、计算设备及存储介质（公开）
- CN107851041A – 多处理器/多核心计算系统的动态调优（权利人：英特尔）
- CN110741351B – 确定虚拟化多处理系统的处理器利用率（授权）
- CN107851041B – 多处理器/多核心计算系统的动态调优（授权）
- CN110998539A – 系统更新的性能影响分析（公开）
- CN105164651B – 在管理的运行时间环境域中的高速缓存管理（授权）
🇨🇳 中国专利申请（已受理 / 实审中）
- 一种面向混合架构的CPU利用率的计算系统和方法. 发明人：周经森、江新宇、冯雨森、管江涛. 状态：实审中. 申请日：2023.11
- 一种基于机器学习的数据库性能预测方法. 发明人：周经森、孙志超. 状态：实审中. 申请日：2024.11.20
- 一种面向电商秒杀应用的基准测试方法. 发明人：周经森、陈奕坤、杨孟铎、常亚辰、江新宇、章超. 状态：将要授权. 申请日：2024.10.31；预计授权日：2025.10.20
- 一种基于类别感知和特征解耦的分布外检测方法. 发明人：周经森、常亚辰、凌志威、赵海亮. 状态：实审中. 申请日：2025.01.23
- 一种云服务器异常检测方法. 发明人：周经森、梁冬晴. 状态：实审中. 申请日：2025.01.22
- 一种自动提取并行应用程序热点代码的方法. 发明人：周经森、章超. 状态：将要授权. 申请日：2024.12.09；预计授权日：2025.09.26
- 一种多个核心组内共享预取器的预取配置优化方法. 发明人：周经森、常亚辰. 状态：将要授权. 申请日：2024.12.10；预计授权日：2025.09.25
- 一种面向数据中心集群的多重连接聚类方法. 发明人：周经森、冯雨森. 状态：将要授权. 申请日：2024.12.06；预计授权日：2025.09.22
- 一种CPU性能采样工具的运行开销的预测方法. 发明人：周经森、汤煜. 状态：受理. 申请日：2025.07.08
- 一种基于分布外检测的联邦学习方法. 发明人：周经森、章超、赵海亮、凌志威. 状态：受理. 申请日：2025.04.01
- 一种计算机处理器性能监测单元的硬件事件组调度方法. 发明人：周经森、江新宇. 状态：受理. 申请日：2025.07.08
- 一种基于LLM聚类和多次召回的文档检索方法. 发明人：周经森、管江涛. 状态：受理. 申请日：2025.10.11
🌐 International Applications（5）
📮 Contact
- LinkedIn: Kingsum Chow
- Email: ksumchow@outlook.com
- Location: Ningbo, Zhejiang, China