large scale deployment of dap and dts rob kooper jay alemeda volodymyr kindratenko

30
LARGE SCALE DEPLOYMENT OF DAP AND DTS Rob Kooper Jay Alemeda Volodymyr Kindratenko

Upload: ernest-stewart

Post on 27-Dec-2015

218 views

Category:

Documents


2 download

TRANSCRIPT

PowerPoint Presentation

Large scale deployment of DAP and DTS

Rob KooperJay AlemedaVolodymyr KindratenkoThe need for scalingHow can we scale?How can DAP architecture scale?How can DTS architecture scale?

What options do we have to scale?Amazon solution for scalingXSEDE solution for scalingCloud solution for scaling

Finite ResourcesScalability A system whose performance improves after adding hardware, proportionally to the capacity added, is said to be a scalable system.

Scaling Up And OutScale UP (vertically)Adding resources to a single systemSpeedPerformanceMoors LawScale OUT (horizontally)CloudAdding nodes to the systemNodes can be commodity hardware (vs HPC)Increase software complexityIncrease management complexityElasticityNeed ability to grow/shrink on demandBased on workload add or remove resources

Keep requirements smallIf many people use one service bring up more of thoseDont bring up services that people dont use

Software Server ArchitectureSoftware ServerSoftware ServerSoftware ServerSoftware ServerImageMagicOpen Officeffmepg3D StudioPolyglotUnknown FormatDataUseableDataMedici 2.0 ArchitectureFrontendWebappLoad BalancerMongoDBHTTPHTMLJSONHTTPHTML JSONExternalServicesFrontendWebappFrontendWebappEvent Bus (rabbitMQ)Extractor(Java)Extractor(Python)ElasticsearchFilesystemMongoDBMongoDBElasticsearchElasticsearch8How to grow?More servers at ISDAFunding is in Brown DogNot sustainable

Commercial CloudsAmazon,

XSEDENSF funded HPC computation

NCSACloud infrastructureAWS Web Application Reference Architecture

AWS Batch Processing Reference Architecture

11PricingSmall machine (1CPU, 2GB)Linux $0.026 per HourWindows $0.036 per Hour

Server is approx. $10,000 and can hold 20 VMsAverage lifespan 5 years (~ $500 per VM)Equals around 2 years of Amazon time

But cheaper if we only need it 8 hours per day! And 7 hours/day in case of windows.XSEDE ResourcesJay AlamedaNational Center for Supercomputing Applications23 July 2014What is XSEDEIntegrating service for wide variety of High Performance Computing (HPC) and Visualization and Data Analysis (RDAV) resourcesFront line supportUniform documentationExtended collaborative supportTraining, education and outreach servicesAllocationswww.xsede.org

Variety of HPC and RDAV resourcesDynamic list at https://www.xsede.org/web/guest/resources/overviewOverview, and expiration dates for each resourceTraditional clustersVisualization and data analysis resourcesStorage resourcesHigh throughput resourcesTestbedsServices

Potentially Interesting Resources for BrowndogTestbed resource FutureGridProduction through 9/30/2014Partitioned intoHPCInfrastructure as a Service (IaaS)NimbusOpenstackEucalyptusDedicatedLayer Platform as a Service (PaaS) (eg, MapReduce, Hadoop) on top of these partitions

Potentially Interesting Resources for Browndog - 2Service resource QuarryWeb service hosting environmentResource end date not specifiedAvailable for XRAC allocations with web-service componentStorage: either NSF home directories, or lustre based storage.OpenVZ provides virtual hosting of RPM based linux distributionsPersistent virtual machineNew XSEDE Resource: CometLong-tail science system hosted at San Diego Supercomputer CenterBuilds on experience with SDSC Gordon (flash memory, persistent storage nodes), and SDSC Trestles (long-tail science)99% of jobs in 2012 used < 2048 coresThese jobs consumed half of the total core hours across NSF resources.CometPartially designed to pick up FutureGrid use (virtual clusters)Gateway hosting nodes and virtual machine repositoryOptimized for jobs within a rackContinues access to flash memory (Gordon)Capacity computing: computing for the 99% of XSEDE jobsComet virtualizationLeverage experience and expertise from FutureGridVirtual machine jobs scheduled like batch jobsFlexible software environments for new communities and applicationsVirtual machine repositoryVirtual HPC cluster (multi-(whole)-node), miminum latency and overhead penaltyXSEDE and BrownDogPremise: BrownDog will become an integral part of a researchers workflowQuestion: Should BrownDog evolve into an XSEDE resource provider, to provide data services for XSEDE?

ISL ResourcesVolodymyr KindratenkoInnovative Systems LaboratoryNational Center for Supercomputing ApplicationsUniversity of Illinois at Urbana-ChampaignHadoop

OpenStack Cloud

Virtual Lab for Advanced Design

http://www.ncsa.illinois.edu/about/org/islHigh memory nodeDell PowerEdge R920

CPUIntel Xeon E7-4860v2 2.6 GHz (4)RAM3 TBStorage2x 300 GB 10,000 RPM SAS 6 Gbps HDD4x 800 GB SAS Read-Intensive MLC 12 Gbps SSD6x 1 TB 7,200 RPM Near-Line SAS 6 Gbps HDD Interconnect6x 1 Gbps Ethernet2x 10 Gbps Ethernet

CPU0CPU1CPU2CPU3RAMRAMRAMRAMRAMRAMRAMRAMPHCQPIPCIe, DMAOther systemsGPU Server8 NVIDIA C2050 GPUsIntel Xeon Phi Server2 Xeon Phi 7120 (Knights Corner) application acceleratorsHPC cluster8 nodes