## GPU Resource Management On JDOS 梁永清 liangyongqing1@jd.com ## 提供的服务 ## Experiment ## Training 1. 用于实验的 GPU 容器 2. 基于 Kubeflow 的机器学习训练服务 3. 模型管理和模型 Serving 服务 ## Serving 均基于容器，不对业务方直接提供 GPU 物理机

## +24 ## Back To Basics Lifetime Management ## PHIL NASH ## 20 24 September 15 - 20 ## C++ is complex  mostly for historical sonarsource.com/blog/beyond-the-rules-of-three-five-and-zero/ ## +24 ## Back To Basics Lifetime Management ## PHIL NASH ## 20 24 September 15 - 20

24. June # Secrets Management at Scale with Vault & Rancher  Bastian Hofman Senior Field Engineer SUSE bastian.hofmann@suse Operations & Infrastructure Management (Run & Manage) K8s Version Management GitOps Continuous Delivery Cluster Templates & Config Enforcement Node Pool Management RBAC, OPA, Pod & Network Network Policies Cluster Provisioning & Lifecycle Management  kubernetes ![Image](/uploads/documents/4/7/a/c/47acf68

## +23 ### BehaviorTree.CPP: Task Planning for Robots and Virtual Agents ## DAVIDE FACONTI ## 20 23 October 01 - 06 ## What you will learn today 1. What Behavior Trees are 2. About the C++ library ments/8/d/b/6/8db662e9d441565a9dd326035c73ae34/p7_1.jpg) ## Robot Behaviors This Coordinator or Task Planner is also the highest level of abstraction of our system, where we focus on WHAT the robot does

## Node Operator: Kubernetes Node Management Made Simple 陈俊(Joe)，Ant Financial ## Agenda • Background and Motivation • Introduction of Operators • Node-Operator • Advanced Topic: Kube-on-Kube-Operator Master & Node Components reliably • Canary Rollout • Master & Node Component Versions Management  Worker Order Complicated architecture Work order deployment system can not meet the requirements of resource management. ## Operator 

## +24 ## Work Contracts Rethinking Task Based Concurrency and Parallelism for Low Latency C++ MICHAEL A MANISCALCO ### “We cannot solve our problems with the same thinking we used when we created them So what is there to Rethink? ## Rethinking: Task Queues  ## Problem #1 - Task Queues Do Not Scale Well: • Contention: ☐ problems: ☐ Task starvation ☐ Load balancing ☐ Forfeits strict FIFO behaviour ○ Increases memory footprint (or requires allocations) ☐ Terrible task selection "fairness" ## Rethinking: Task Queues

## +23 ## Taro: Task graph-based Asynchronous Programming Using C++ Coroutine DIAN-LUN LIN 20 23 October 01 - 06 ## Agenda • Understand the motivation behind Taro • Learn to use the Taro C++ programming application • Conclusion ## What is Task Graph-based Programming System (TGPS) • TGPS encapsulates function calls and their dependencies in a top-down task graph ## What is Task Graph-based Programming System function calls and their dependencies in a top-down task graph  ## What is Task Graph-based Programming System (TGPS) • TGPS encapsulates

Toward Standard C++ Dependency Management ## BILL HOFFMAN & BRET BROWN ## 20 23 October 01 - 06 ## Libraries: A First Step Toward Standard C++ Dependency Management October 3, 2023 Bloomberg Engineering portable as the code they contain! ✓ Projects should be “cattle,” not “pets”! ## Why dependency management? Consensus: Managing dependencies == way too hard Q: Which of these do you find frustrating about following names: jsonlogConfig.cmake Jsonlog-config.cmake # ... CMake gives you some dependency management tips here ... Aside: Coloring and bolding added for emphasis ## Motivation: What would we design

or DND  • No non-task related browsing ☐ No bathroom breaks $ ^{*} $ • No conversations ![Image](/uploads/documen No eating • No preening ## How does it work? 1. Choose a task you’d like to get done 2. Set the Pomodoro for 25 minutes 3. Work on the task until the Pomodoro rings 4. When the Pomodoro rings, put activity requires you by monitoring how many Pomodoros you need in order to accomplish a specific task 2. Learn to protect your Pomodoro from the internal and external interruption 3. Make accurate estimations

# CS 591 K1: Data Stream Processing and Analytics Spring 2020 2/25: State Management Vasiliki (Vasia) Kalavri vkalavri@bu.edu ## State in dataflow computations Any non-trivial streaming computation sum, list, map, ... ## State management in Apache Flink All data maintained by a task and used to compute results: a local or instance variable that is accessed by a task’s business logic Operator state state is scoped to an operator task, i.e. records processed by the same parallel task have access to the same state - It cannot be accessed by other parallel tasks of the same or different operators Keyed