Performance of Apache Ozone on NVMe Wei-Chiu Chuang (jojochuang) Ritesh Shukla (kerneltime) Agenda • Overview of how Ozone and how it scales • Why NVME is important for Ozone for scaling • Benefits Benefits of using NVME • Impala performance results from NVME clusters • Write path improvements results from NVME clusters • Summary • Questions Ozone Architecture Why does Ozone Scale? Separation of restart) Slow startup based on size High Availability Active - Active Active - Standby Protocol Support Hadoop / S3 API Hadoop API Small objects are welcome Max performance reached at object size around

#IstioCon Performance tuning and best practices in a Knative based, large-scale serverless platform with Istio 张龚, Gong Zhang, IBM China Development Lab 庄宇, Yu Zhuang, IBM China Development Lab #IstioCon Cloud. #IstioCon ● Knative and Istio ● How Istio is leveraged in a Knative based platform ● Performance bottleneck analysis and tuning ○ Istio scalability optimization during Knative Service provisioning mesh enabled (based on https://github.com/knative/serving) #IstioCon Performance bottleneck analysis and tuning • Performance Criteria: the platform has multiple shard k8s clusters, each cluster should

Kalavri
 vkalavri@bu.edu Spring 2020 3/17: High availability, recovery semantics, and guarantees Vasiliki Kalavri | Boston University 2020 Today’s topics • High-availability and fault-tolerance in distributed Kalavri | Boston University 2020 Fault-tolerance trade-offs 12 Steady-state overhead • How is performance affected by the fault-tolerance mechanism under normal, failure- free operation? • How much state consists of • input queues • operator state • output queues • Short recovery time • High runtime overhead • The checkpoint interval determines the trade-off 14 Ni primary secondary

Data Stream Processing and Analytics Vasiliki (Vasia) Kalavri
 vkalavri@bu.edu Spring 2020 3/31: High-availability & reconfiguration ??? Vasiliki Kalavri | Boston University 2020 • To recover from failures recover its state 2 Checkpointing guards the state from failures, but what about process failure? High-availability ??? Vasiliki Kalavri | Boston University 2020 3 Flink processes ??? Vasiliki Kalavri • A high-availability mode migrates the responsibility and metadata for a job to another JobManager in case the original JobManager disappears. • Flink relies on Apache ZooKeeper for high-availability

由集群中运行的用户应用程序生成的容器日志（基础架构容器应用程序除外）。 infrastructure - 在集群和 OpenShift Container Platform 节点上运行的基础架构组件生成的日 志，如 journal 日志。基础架构组件是在 openshift*、kube* 或 default 项目中运行的 pod。 audit - 由 auditd 生成的日志，节点审计系统存储在 /var/log/audit/audit 描述。Elasticsearch 无法使用原始块卷。 OpenShift Elasticsearch Operator 使用 Elasticsearch 资源名称为 PVC 命名。 Fluentd 将 systemd journal 和 /var/log/containers/ 的所有日志都传输到 Elasticsearch。 Elasticsearch 需要足够内存来执行大型合并操作。如果没有足够的内存，它将会变得无响应。要避免这个 输出示例 出示例 片刻后，原始 Kibana Pod 将被删除。 输 输出示例 出示例 4.9. 配置 SYSTEMD-JOURNALD 和 FLUENTD Fluentd 需要从日志 (journal) 中读取数据。因为日志默认设置非常低，它可能无法跟上系统服务的日志记 录率，所以日志条目可能会丢失。 我们推荐设置 RateLimitIntervalSec=30s 和 RateLimitBurst=10000

Message Messa e Filter Server Triggered Redelivery Message Storage Message Retroactive essage Priority High Availability and Failover Message Track Configuration anagement and Operation Tools ActiveMQ Java Supported Supported Not Supported Supports very fast p rsistence using JDBC along with a high performance journal，such as levelDB, kahaDB S pported Supported Supported, depending on storage,if using kahadb async producer Not Supported upported, you can use Kafka Streams to filter messages Not Supported High perfo mance file storage Supported offset indicate Not Supported Supported, requi es a ZooKeeper

results. For example, between quantization and clustering, which one is preferable? What is the performance impact when both are used together? We have four options: none, quantization, clustering, and both past few years, we have seen newer architectures, techniques and training procedures pushing the performance benchmarks higher. Figure 7-1 shows some of the choices we face when working on a deep learning process of learning are called hyperparameters to differentiate them from model parameters. The performance of deep learning relies on a set of good hyperparameters. Some of the commonly tuned hyperparameters

With the Main API, you can create, configure, start, stop and delete virtual machines, retrieve performance statistics about running VMs, configure the VirtualBox installation in general, and xxvii Introduction managed object references. As a result, this timeout value should not be set too high, especially on machines with a high load on the web service, or the web service may eventually deny service. • --check-interval the same “Main API”: the web service or COM/XPCOM, page xxviii, the COM/XPCOM API gives better performance without the SOAP overhead, and does not require a web server to be running. On the other hand,

With the Main API, you can create, configure, start, stop and delete virtual machines, retrieve performance statistics about running VMs, configure the VirtualBox installation in general, and 1 1 Introduction managed object references. As a result, this timeout value should not be set too high, especially on machines with a high load on the web service, or the web service may eventually deny service. • --check-interval guises of the same “Main API”: the web service or COM/XPCOM, page 2, the COM/XPCOM API gives better performance without the SOAP overhead, and does not require a web server to be running. On the other hand,

With the Main API, you can create, configure, start, stop and delete virtual machines, retrieve performance statistics about running VMs, configure the VirtualBox installation in general, and 27 1 Introduction managed object references. As a result, this timeout value should not be set too high, especially on machines with a high load on the web service, or the web service may eventually deny service. • --check-interval of the same “Main API”: the web service or COM/XPCOM, page 28, the COM/XPCOM API gives better performance without the SOAP overhead, and does not require a web server to be running. On the other hand,