WIXLIB

Reference ArchitectureMirantis OpenStackReference Architecturefor Dell HardwareOverviewThis document describes the Mirantis OpenStack for Cloud Native Appswith Dell—a fully validated deployment of Mirantis OpenStack on Dell PowerEdge R630 and R730xd servers networked with Dell NetworkingS3048-ON and S4048-ON switches.The deployment is engineered as a scalable, rack-based OpenStack dev/test/production environment for cloud-native web applications.This Reference Architecture details all hardware and software componentsof the Mirantis OpenStack for Cloud Native Apps; describes their physicalintegration and logical interoperation; and discusses in depth certain criticalaspects of the design relevant to performance, high availability, data-lossprevention, and scaling.Use Case SummaryThis Reference Architecture (RA) is designed for cloud-native applications.It comes with the Murano catalog preloaded to support deployment ofdeveloper tools, databases, CI/CD toolchain elements (e.g., Jenkins),leading PaaS solutions (e.g., Cloud Foundry), containers and orchestrators(e.g., Docker, Kubernetes), and other solutions, including those available inthe online OpenStack Community App Catalog. Neutron networking allowsfor self-service networking isolated by tenants while Ceph provides bothblock and object storage to support various application types.

REFERENCE ARCHITECTURE2Table of ContentsMirantis OpenStackOverview . . . . . . . . . . . . . . . . . . . . . . . . 1Mirantis is a number one contributorto OpenStack. Fuel—a core part of theMirantis OpenStack distribution — hasbeen taken under Big Tent.Use Case Summary. . . . . . . . . . . . . . . 1Mirantis OpenStack . . . . . . . . . . . . . . 2Networking. . . . . . . . . . . . . . . . . . . . 2Mirantis OpenStack Architecture. . 3Nodes Types. . . . . . . . . . . . . . . . . . . . . 3High Availability. . . . . . . . . . . . . . 4Mirantis recommends the followingconfiguration of Mirantis OpenStack: Mirantis OpenStack version 7.0,based on OpenStack Kilo.Optional Components. . . . . . . . 4 Ubuntu 14.04 as host OS.Network Architecture. . . . . . . . . . . . . 5 The following OpenStack servicesdeployed: Keystone, Nova, Glance,Cinder, Neutron, Horizon, Heat,Murano, Ceilometer.Rack Physical Cabling Schema. . . . 7Server Interface Configuration. . 8Logical Networking Schema. . . . . . . 9Rack Hardware Specification. . . . . 10Disk Configuration. . . . . . . . . . . . . . . 11Validated Configuration . . . . . . . 12Scaling . . . . . . . . . . . . . . . . . . . . . . . 12Suggested Rack Configurations . . 13Testing . . . . . . . . . . . . . . . . . . . . . . . 13References. . . . . . . . . . . . . . . . . . . . . . 14Glossary. . . . . . . . . . . . . . . . . . . . . . . . 14Trademarks. . . . . . . . . . . . . . . . . . . . . 15 Three OpenStack Controller nodes inan HA configuration usingHAProxy, Corosync/Pacemaker,MySQL Galera, RabbitMQ, MongoDB. A Ceph cluster provides a redundantbackend for Glance, Cinder, NovaEphemeral, and Swift-API.– Ephemeral volumes configuredto use Ceph backend to providelarge volumes and support livemigration– 3x replication is utilized toprotect data Secure public OpenStack APIendpoints and Horizon with TLSv1.2.24 DIMMS of DDR4 RAM, and supportthe 12-Gb PowerEdge RAID controllers(PERC9). These features, combined witha high-performance energy-efficientdesign, should serve well for a varietyof workloads your cloud infrastructurewill need to support.The server configurationsrecommended are well-tested andproven with the Mirantis OpenStackfor Cloud Native Apps referencedesign, creating a foundation that cansecure your infrastructure investmentand allow for easy expansion of yourgrowing cloud.NetworkingThe network layer is based on Neutronwith OpenvSwitch and VLANs fortenant traffic separation because thisconfiguration eliminates the cost of anSDN solution.10-GbE networking infrastructure plusthe Neutron DVR feature enables highperformance East-West traffic.Designated NICs used for Storagereplication traffic reduce pressurefrom tenant traffic. NIC bonding andmulti-chassis LAG across multipleswitches for uplinks and downlinksprovide both high performance andphysical redundancy for the wholenetwork infrastructure. Zabbix or LMA toolchain monitoringservices installed.Key Benefits Sahara (optional). E asy to deploy integrated solution byMirantis on Dell PowerEdge Servers.Hardware OptionsFor this Reference Architecture, the DellPowerEdge R630 and R730xd serverswere chosen because of their hardwareoptions, performance, and scalability.Both platforms are powered by theIntel Xeon processor E5-2600 v3product family, are able to utilize up to F lexible enough to cover a broad setof use cases. Based on best-of-breed,enterprise-ready Mirantis OpenStackand Dell PowerEdge Servers.

REFERENCE ARCHITECTURE3Mirantis OpenStack ArchitectureMirantis OpenStackMirantis OpenStack consistsof the following “core” OpenStackservices: Keystone Nova Neutron Cinder GlanceSwift is substituted by Ceph withRadosGW providing the sameSwift API.The following “optional”OpenStack services are installedin Mirantis OpenStack: Horizon Heat Murano CeilometerAll services except Ceilometer areusing local MySQL as a databasebackend. Ceilometer usesMongoDB. All services are usingRabbitMQ as a messaging queueservice. Ceph provides Swift APIvia RadosGW and also is used asa backend for Cinder and Glance.Mirantis OpenStack Overview.Mirantis OpenStack 7.0 is aKilo-based OpenStack distributionrunning on top of Ubuntu 14.04.Node TypesServers used in this ReferenceArchitecture will serve as one ofthese node types: Infrastructure,Controller, Compute, or Storage.which incorporates all coreOpenStack infrastructure servicessuch as MySQL, RabbitMQ,HAProxy, OpenStack APIs, Horizon,and MongoDB. This node is notused to run VMs.Infrastructure NodeCompute NodeThe Infrastructure node is an Ubuntu14.04-based node, which carries twovirtual appliances:The Compute node is a hypervisorcomponent of a cloud, which runsvirtual instances. Fuel Master node —an OpenStackdeployment tool.Storage Node Cloud Validation node —a setof OpenStack post-deploymentvalidation tools including Tempestand Rally.Controller NodeThe Controller node is a controlplane component of a cloud,The storage node is a componentof an OpenStack environment,which keeps and replicates all userdata stored in your cloud includingobject and block storage. Ceph isused as a storage backend.Nova Ephemeral storage may uselocal disks or Ceph. The local diskgives you better performancebut Ceph enables live migrationfor instances with Ephemeralstorage used. Placement ofEphemeral storage to localdisks is considered as technicalpreview.To learn more, please see thedescription of the MirantisOpenStack architecture 0/reference-architecture.html.

REFERENCE ARCHITECTURE4High AvailabilityThe High Availability modelimplemented in Mirantis OpenStack(MOS) is well described in officialdocumentation adeploymentHere is the short description.To protect core OpenStack servicesfrom failure of a Controller node,control plane components areduplicated on multiple Controllernodes. It’s possible to deploy 3, 5, 7,or any other odd number of Controllernodes and add more nodes afterthe initial deployment (but keepingodd total number) to distribute loadand increase the redundancy level.Pacemaker/Corosync is used to managecritical infrastructure parts when failureoccurs. Because of the quorum majorityvoting algorithm used in Corosync,a cluster of ‘N’ Controller nodes maysurvive when ‘N/2-1’ nodes go down.For three nodes it’s one node down,for five nodes it’s two nodes, and so on.Almost all services are working inactive/active mode and load is balancedby HAProxy listening at a virtual IPaddress. OpenStack componentssupport active/active mode natively,as do RabbitMQ and MongoDB. Galeracluster is used to protect MySQL.From the networking perspective, highavailability is upheld by redundantand bonded connections to two ToRswitches. That means if one cable getscut or an entire switch goes down,a cloud will survive seamlessly withhalf its normal network bandwidthstill available.Another low-level hardware feature thatprovides redundancy is RAID1 for diskson Infrastructure and Storage nodesto keep an operating system intact incase of a disk failure. Also it’s highlyrecommended to use two independenthot-plug power supplies for eachhardware piece. This helps protectcomponents from powersupply failures.Ceph provides redundancy for storeddata. The recommended replicationnumber is three, which means at a giventime there are three copies of the samedata spread across Ceph nodes.a separate physical disk on each node.Each Controller node contains twoSSDs: the first one for OS plus otherservices and the second onefor MongoDB only.Optional ComponentsSahara is a service used to deploybig data (i.e., Hadoop) clusters on topof OpenStack that may be installedon demand. Sahara wasn’t used duringthe validation.There are two possible ways to useMongoDB with Ceilometer: installthe database locally or externally.Both options are available in MirantisOpenStack. When installing MongoDBlocally we recommend using at leastthree separate servers to form aMongoDB cluster. Placing MongoDBonto Controller nodes may causeresource starvation for key servicessuch as MySQL or RabbitMQ, which maylead to severe issues with the cloud asa whole. The best approach is to usededicated nodes for MongoDB; eitherexternal ones or nodes deployed aspart of MOS.For validation purposes, we putMongoDB on the Controllers, but give itBy default, Mirantis OpenStack protectspublic OpenStack APIs and Horizonwith TLSv1.2 using self-signed or userprovided certificates, but it’s possibleto disable this feature if necessary. Forvalidation purposes, we used selfsigned certificates.Mirantis OpenStack may be extendedvia Fuel plugins to provide twovalidated options to monitor your cloud:Zabbix or LMA toolchain (Elasticsearch/Kibana, InfluxDB/Grafana, and Nagios).Neither was used during this validation.