Optimizing Cluster Utilisationwith Bright Cluster ManagerArno ZiebartSales Manager GermanyHPC Advisory Council 2011www.clustervision.com12010

About us Specialists in Compute, Storage & GPU Clusters(Tailor-Made, Turn-Key) Unique position in Europe (EMEA)Oslo Offices in Amsterdam, Gloucester, Munich, Paris, Milan, Geneva, Madrid, Oslo, USA 50 Staff, most technical, all specializedin clustersGloucester Amsterdam Hardware independent Background in Science, Research,ParisMunich EngineeringGeneva At forefront of clustering technologyMilan Award winning (Intermediair,MadridVosko, NBCC, Deloitte Rising Star) Financially strong, profitable, growing Over 300 customerswww.clustervision.com22010

Customers — Governmentwww.clustervision.com32010

Customers — Industrywww.clustervision.com42010

Customers — Academiawww.clustervision.com52010

Customers — TOP500TOP500 (2008/2009/2010) University FrankfurtRU GroningenSaudi Aramco (Saudi Arabia)University of Cambridge (UK)University of Bristol (UK)University College London (UK)CASPUR (Italy)University of Gent (Belgium)www.clustervision.com62010

ClusterVision Customers22nd on TOP500 list (Nov. 2010)20 784 CPU cores (2.1GHz)772 Ati Radeon HD 5870 GPUsFastest x86-based System in GermanyFastest system in the world based on AMD/ATI GPUs60.7% efficiencywww.clustervision.com72010

Customers — TOP500www.clustervision.com82010

Products & Services Turnkey clusters–––Compute clustersStorage clustersGPU clusters Cluster software––Bright Cluster ManagerMS Windows HPC Server 2008 HPC signdeploymentsupportcooling Parallel file systems––––LustreFraunhofer Global Filesystem (FhGFS)IBM GPFS (Official world-wide OEM)NASwww.clustervision.com92010


Cluster Architecture HeadNode01node001x 24PDUsx 16Switchnode002MonitoringNode01node003SNMP x8RacksMonitoringNode02node511ProvisioningNode01 e5122010

Cluster Architecture HeadNode01node001x 24PDUsx 16Switchnode002MonitoringNode01node003SNMP x8RacksMonitoringNode02Cluster Managementnode511ProvisioningNode01 e5122010

Bright Cluster Managerwww.clustervision.com182010

Cluster Management Most solutions use the “toolkit” approach Tools typically used: Ganglia, Cacti, Nagios,Cfengine, xcat, etc Issues with the “toolkit” approach: Tools rarely designed to work togetherTools rarely designed for HPCTools rarely designed to scaleEach tool has its own command line interface and GUIEach tool has its own daemon and databaseRoadmap dependent on developers of the tools Making a collection of unrelated tools work together Requires a lot of expertise and scriptingRarely leads to a really easy-to-use and scalable solutionOften leads to long installation and ramp uptimeLow throughputwww.clustervision.com192010

Cluster Management Bright Cluster Manager takes a much morefundamental & integrated approach Designed and written from the ground upSingle cluster management daemon provides all functionalitySingle, central database for configuration and monitoring dataSingle CLI and GUI for ALL cluster management functionality Which makes Bright Cluster Manager Extremely easy to useExtremely scalableSecure & 0

Bright Integrated ApproachUtilisationBetter throughputand UtilisationFaster timeto full UserProductivityFaster timeto full systemreadinessStrong Policies drivenallocation of resources“Sweat theAssets” muchearlierTime in months 21www.clustervision.com212010

Bright Cluster ManagerCluster ManagerMolecular Intel Cluster Ready certifiedBiophysic Integration and SupportQCDPhysicsCFDChemicalCluster Manager Years of HPC expertise User Moduls Environment Cluster AdministrationCluster Administration MonitoringParallelFilesystem Node boot and provisioning system Linux onitoringWorkload ManagementAccount.HPC User Environment Workloadmanager HPC MiddlewareManufacturingRedhat2010

Management InterfaceGraphical User Interface (GUI) Offers administrator full cluster controlStandalone desktop applicationManages multiple clusters simultaneouslyRuns on Linux, Windows, MacOS XBuilt on top of Mozilla XUL engineAdmin GUICommand Line Interface (CLI) All GUI functionality also availablethrough Command Line Interface (CLI)Interactive and scriptable in batch modewww.clustervision.com24Admin CLI2010

Architecture — MonitoringClusterCMDaemonAdmin GUImetricsnode001eventsmonitoring datametricsmetricsmonitoringdataHead Nodenode002monitoring datametricsRaw dataConsolidateddatanode003


Bright GPU Metricswww.clustervision.com462010

Bright GPU Metricswww.clustervision.com472010

Bright Cluster ManagerAdvanced Features Daemon with low resource consumptionSynchronised daemon to prevent OS jitterMultiple, load-balanced provisioning nodesNode discovery using Ethernet switch port detectionLive & incremental image updatesAutomated BIOS updates and configurationsInfiniband only storage & diskless client supportNode and service checks (pre/post to scheduler)Roadmap Features More power saving features Virtualisation, Cloud computingwww.clustervision.com532010

Professional Services Application AnalysisCluster designBenchmarkingCluster installation Training Hardware support––––Collect and ReturnOnsite RepairNext Business Day Repair4 Hour Response 24 x 7 Software support– Bright Cluster Manager (Bright)– System administration (onsite/remote)– User helpdesk Code porting, optimisation & parallelizationwww.clustervision.com542010

Conclusions Proven track-record in cluster computing Hardware independent, partnering with Dell Best cluster software stack on the market––––Easy manage and useSuitable for very large clustersComprehensive HPC user environmentComplete & consistently integrated 100% committed to cluster computingwww.clustervision.com632010


The Endwww.clustervision.com652010

- Bright Cluster Manager - MS Windows HPC Server 2008 HPC Services - Cluster design - Cluster deployment - Cluster support - Cluster cooling Parallel file systems - Lustre - Fraunhofer Global Filesystem (FhGFS) - IBM GPFS (Official world-wide OEM) - NAS .