Rook-EdgeFS技术深度剖析课件.pptx

1、Deep dive: Rook / EdgeFSRook EdgeFS技术深度剖析What is Rook? Cloud-Native Storage Orchestrator Extends Kubernetes with custom types and controllers Automates deployment, bootstrapping, configuration, provisioning, scaling, upgrading, migration, disaster recovery, monitoring, and resource management Framew

2、ork for many storage providers and solutions Open Source (Apache 2.0) Hosted by the Cloud-Native Computing Foundation (CNCF)Storage ON Kubernetes Deploy storage systems INTO the cluster Harness the power of Kubernetes Automated management by smart software Portable abstractions for all our storage n

3、eedsRook is a FrameworkRook is more than just a collection of Operators and CRDsFramework for storage providers to integrate their solutions into cloud-native environmentsStorage resource normalizationOperator patterns/plumbingCommon policies, specs, logicTesting effortEdgeFS, Ceph, CockroachDB, Min

4、io, NFS, Cassandra, and moreNew Rook OperatorsApache CassandraNexenta EdgeFSRook EdgeFS Deployed as Kubernetes Operator Full service life-cycle management: install, update, rolling-upgrade, uninstall, re-install Monitoring with Prometheus and Grafana Easy of use built-in GUI with CRD Wizard! Runs in

5、 Embedded environments: 1GB DRAM, 2 CPU cores! Consumes Locally connected raw disks or directories Provides Globally available Data protocols S3 and S3X (NoSQL and POSIX-like extension) Scale-Out High-Performance NFS Scale-Out iSCSI Block DevicesEdgeFS Connected Data Global Deduplication Clouds Conn

6、ectivity Geo-Transparency Global Namespace Protocol Transparency S3, S3X, NFS, iSCSISegment 1MySpace/WorkHDDNSSD0HDD0ISGW LinkNFS GWS3 GWSegment NMySpace/HomeHDDNSSD0HDD0ISGW LinkNFS GWS3 GW* Block-level Geo-Deduplication* Metadata-Only Transfers* Local Caching* Intelligent PrefetchingUp to 50 x ban

7、dwidth savings!EdgeFS Cloud Connectors Presents S3 Everywhere, Transparently syncing regions! Metadata-Only synchronization with Local Caching and Prefetching. Globally de-duplicates cross-site calls. Operates with unmodified, native objects Supported connectors as of EdgeFS v1.2 AWS S3: Stable Alib

8、aba OSS: Beta GCP GS: Beta Azure Blob: AlphaEdgeFS Multi-Cloud Layer Geo-ScalabilitySpans a network of geographically distributed sites, connected as one global namespaceGit-like architecture with fault-tolerance and immutable, versioned metadata designScales equally well for Object, File, Built-In

9、NoSQL or Block devices Geo-TransparencyAlways ON, bi-directional access to same S3 bucket, NFS exportAutomatic “Last-Writer-Wins” update strategy for S3, Snapview Groups for NFS/iSCSI Geo-ConsistencySnapview groups “floating” within connected geo-namespaceAny granularity of protection: Files, Direct

10、ories, Buckets, LUNs, NoSQL Databases Geo-Locality and Active E-LRU CachingMetadata is always replicated, data is prefetched on-demand and cached locally Modifications synchronized asynchronously, thus geographically eventualEdgeFS Use Cases Multi-Cloud CDN WorkflowEfficient distribution of content

11、with advanced local caching featuresAvoid full replication, with optional pin/unpin/clone of locally cached contentSynchronization of primary source content on AWS, Azure, Alibaba, GCP and others Cloud High-AvailabilityAutomatic failover of failed cloud links to redundant dataset in a different regi

12、onOperate in offline mode for up to 7 days, synchronizing eventually Edge/IoT To or From CloudCapture edge data in local cache and private clouds for AI/ML processingImprove link utilization by sending de-duplicated data asynchronouslyAccess global namespace transparently, while avoiding the need to

13、 do full replication Kubernetes Persistent Volumes across cloudsBi-directional PVs with geo-transparent synchronization across regionsCSI managed File or Block PVs, Consistency groupsData segmentation and region awarenessRook Networking1G, 10G, 25G, 40G, 100G Great.But, it is sharing same Pod networ

14、kwith the converged Apps?Not anymore!Multi-Homed NetworkSelecting Multus-CNIBacked by Intel, leading the spaceFlexible selection of SDNsEnables Namespace IsolationFuture alternatives on the RoadmapCNI-Genie KnitterMultus NetworkAttachmentDefinition for Backend:Multi-Homed EdgeFSPod Switch Network, M

15、TU=9000Backend Switch Network, MTU=9000Storage PodStorage PodS3XStorage PodS3XStorage PodBackend IsolationFrontend IsolationAppPersistent VolumeAppIngressAppPersistent VolumeAppEgressPod Switch Network, MTU=9000Storage PodStorage PodS3XStorage PodS3XStorage PodOne Flat Pod NetworkAppPersistent Volum

16、eAppIngressAppPersistent VolumeAppEgressBenefits of transitioning to Isolated Multi-Homed Network:Improved Performance characteristicsImproved Data SecurityImproved QoS and SLAS3X DeploymentDemo SetupTarget Pod 0 (4TB)HDDSSDHDDHDDHDDTarget Pod 1 (4TB)HDDSSDHDDHDDHDDTarget Pod 2 (4TB)HDDSSDHDDHDDHDDT

17、arget Pod 3 (4TB)HDDSSDHDDHDDHDDS3X COS BenchWorkerHost1Host2Host3Host4Physical Characteristics:4-nodes Kubernetes cluster v1.14, 1 SSD per 4 HDD, 15% SSD of total HDD capacity, 64GB RAM per Host, CPU E5-2630, 2.2GHz10G Pod Network, 10G Backend Network, both Jumbo MTURook S3X Deployment and COS Benc

18、h work generator runs on separate hosts Rook EdgeFS Cluster CRD:Performance Analysis 25-29% better response time! 39-43% better bandwidth!Test was designed to compare Pod vs. Multi-Homed NetworksSingle 10G connected multi-threaded app, 2MB S3 object transfers, 3 sync replicas, 6TB random I/O datasetSummary Rook Community is growing! Top Contributors: Upbound.io, RedHat, Nexenta/DDN, SUSE, Cloudability, more is coming! EdgeFS is emerging to address Multi-Cloud and Edge/IoT needs Multi-Homed network is going to be available in upcoming 1.1 releaseThank you!https:/