What’s new in Apache Ozone 1.3 陈怡 Apache Ozone PMC 主席 3 ⽬录 I. Ozone 构架 II. Ozone 1.3 新功能 III. 未来展望 4 Ozone 构架 5 Ozone 1.3 新功能 I. 纠删码(Erasure coding) II. 系统均衡器(Container Balancer) III

Monitoring & troubleshooting ● Sharing with non-technical stakeholders ● Infrequent tasks ● Onboarding new K8s users / learning Kubectl Strengths: ● In-Terminal workflows ● Frequently-repeated tasks ● ● Scripting & automation ● Sharing workflows / reproducibility ● Customization Onboarding new K8s users https://unsplash.com/ Over 50% of survey takers said that Dashboard is very useful or extremely much much more. github.com/kubernetes/dashboard/releases In-progress work ● Migrating from ng1 to ng2 (#3152) ● Migrating metrics from Heapster to Kubernetes Metrics API (#2986) ● Apps list page

杨立昆（Yann LeCun） 杰弗里·欣顿（Geoffrey Hinton） 本吉奥（ Bengio ） 共同获得了2018年计算机科学的最高奖项 ——ACM图灵奖。 机器学习界的执牛耳者 Andrew Ng 中文名吴恩达，斯坦福大学副教 授，前“百度大脑”的负责人与百 度首席科学家。 6 李航, 现任字节跳动科技有限公司人 工智能实验室总监，北京大学、南京 大学客座教授，IEEE 会士，ACM 杰 ，都有相近的高准确度 。于是诞生了机器学习 界的名言： 成功的机器学习应 用不是拥有最好的 算法，而是拥有最 多的数据！ 数据决定一切 数据大小 准 确 率 77 参考文献 [1] Andrew Ng. Machine Learning[EB/OL]. StanfordUniversity,2014.https://www.coursera.org/course/ml [2] 李航. 统计学习方法[M] Friedman J. The Elements of Statistical Learning[M]. New York: Springer,2001. [5] CHRISTOPHER M. BISHOP. Pattern Recognition and Machine Learning[M]. New York: Springer,2006. [6] Stephen Boyd, Lieven

producer = new DefaultMQProducer("please_rename_unique_ group_name"); producer.setNamesrvAddr("127.0.0.1:9876"); producer.start(); for (int i = 0; i < 9; i++) { try { Message msg = new Message("TopicTest10" 例如笔者喜欢将其放在/opt /application 下。 cp rocketmq-console-ng-1.0.0.jar /opt/application/ Step5：启动 rocketmq-conolse nohup java -jar rocketmq-console-ng-1.0.0.jar & 本文来自『中间件兴趣圈』公众号，仅作技术交流，未授权任何商业行为。 1.3 实战：RocketMQ 本文并不会详细分析 RocketMQ 主从同步的实现细节，如大家对其感兴趣，可以查阅 笔者所著的《RocketMQ 技术内幕》或查看笔者博文：https://blog.csdn.net/prestigedi ng/article/details/79600792 二、提出问题 主，从服务器都在运行过程中，消息消费者是从主拉取消息还是从从拉取？ RocketMQ 主从同步架构中，如果主服务器宕机，从服务器会接管消息消费，此时消

controller automatically detects workload provisioning events such as a new workload being instantiated, network provisioning events such as a new virtual network being created, routing updates from internal and manage Contrail using simplified and familiar DevOps tools and processes without needing to learn a new life cycle management (LCM) paradigm. Benefits of Cloud-Native Contrail Networking • Support a rich 10.42.0.9 rke2-s1 kube-system rke2-ingress-nginx-controller-ng4hg 1/1 Running 0 11h 10.42.2.3 rke2-a2

[1] HOSMER D W, LEMESHOW S, STURDIVANT R X. Applied logistic regression[M]. New Jersey: Wiley New York.2000. [2] Andrew Ng. Machine Learning[EB/OL]. Stanford University,2014. https://www.coursera.org/course/ml Friedman J. The Elements of Statistical Learning[M]. New York: Springer,2001. [5] CHRISTOPHER M. BISHOP. Pattern Recognition and Machine Learning[M]. New York: Springer,2006. [6] Stephen Boyd, Lieven

TOM M MICHELLE. Machine Learning[M]. New York: McGraw-Hill Companies,Inc,1997. [2] Hastie T., Tibshirani R., Friedman J. The Elements of Statistical Learning[M]. New York: Springer,2001. [3] CHRISTOPHER CHRISTOPHER M. BISHOP. Pattern Recognition and Machine Learning[M]. New York: Springer,2006. [4] Zhang H., The optimality of naïve Bayes[C]//Proceedings of the 17th International Florida Artificial Intelligence Intelligence Research Society Conference (FLAIRS), Miami, FL, 562-567, 2004. [5] Ng A. Y. and M. I. Jordan, On discriminative vs. generative classifiers: A comparison of logistic regression and naïve Bayes[C]//

≤ L(µ(t−1), X, Z(t−1)) because the new Z(t) = arg minZ L(µ(t−1), X, Z) When we update µ from µ(t−1) to µ(t) L(µ(t), X, Z(t)) ≤ L(µ(t−1), X, Z(t)) because the new µ(t) = arg minµ L(µ, X, Z(t)) Feng clusters MST-based method: Build a minimum spanning tree from the dissim- ilarity graph, and then make new clusters by breaking the link corre- sponding to the largest dissimilarity Feng Li (SDU) K-Means December containing all the data G Measure the average dissimilarity of i ∈ G to all the other i′ ∈ G dG i = 1 nG � i′∈G di,i′ Remove the most dissimilar data i∗ and put it in its own cluster H i∗ = arg max i∈G

Andrew Ng. Machine Learning[EB/OL]. StanfordUniversity,2014. https://www.coursera.org/course/ml [2] Peter Harrington.机器学习实战[M]. 北京:人民邮电出版社,2013. [3] TOM M MICHELLE. Machine Learning[M]. New York: McGraw-Hill Friedman J. The Elements of Statistical Learning[M]. New York: Springer,2001. [5] CHRISTOPHER M. BISHOP. Pattern Recognition and Machine Learning[M]. New York: Springer,2006. [6] Kohavi R.,Scaling up the

当回溯到根节点时，算法结束，此时保存 的最近邻节点就是最终的最近邻。 遍历完（4,5）的左右叶子节点，发现与当 前最优距离相等，不更新最近邻。 所以（4,4）的最近邻为（4,5）。 25 参考文献 [1] Andrew Ng. Machine Learning[EB/OL]. StanfordUniversity,2014.https://www.coursera.org/course/ml [2] 李航. 统计学习方法[M] Friedman J. The Elements of Statistical Learning[M]. New York: Springer,2001. [6] CHRISTOPHER M. BISHOP. Pattern Recognition and Machine Learning[M]. New York: Springer,2006. [7] Stephen Boyd, Lieven