Top Banner

Click here to load reader

Fault-Tolerant Workflow Scheduling Using Spot Instances on Clouds Deepak Poola, Kotagiri Ramamohanarao, and Rajkumar Buyya Cloud Computing and Distributed

Jan 12, 2016



PowerPoint Presentation

Fault-Tolerant Workflow Scheduling Using Spot Instances on CloudsDeepak Poola, Kotagiri Ramamohanarao, and Rajkumar Buyya

Cloud Computing and Distributed Systems (CLOUDS) LaboratoryDepartment of Computing and Information Systems, The University of Melbourne, Email: [email protected],{kotagiri,rbuyya}

ICCS-2014, Cairns, Australia1Cloud ComputingCloud ComputingOffers resources as a subscription based serviceHighly scalableHighly availableDriven by market principlesDynamically configured and delivered on demandDifferent pricing models

2Benefits of Cloud Computing Scalability or elasticityOn-Demand resource provisioningWide range of resource typesPay-as-you-go modelAttractive cost modelsIllusion of unlimited resourcesCheaper and fast storage facilitiesPlethora of tools for ease of useContent-deliveryMonitoring

NetworkingDeployment and Management

3Spot InstancesStarted by Amazon around December 2009Idle or unused datacenter capacitySpot price is decided in an Auction-like mechanismVaries with time and instance typeVaries between regions and availability zonesbid should be higher than or equal to the spot priceOffers upto 60% cost reductionsWorkflowsScientific workflow systems aim at automating large complex data analysis to make it easier for scientists.Workflows are collection of tasks that are data dependent or control dependent. Workflows can be represented as Directed Acyclic GraphWorkflow scheduling maps tasks to resources whilst maintaining dependenciesJargonsMakespanCost

Sample Workflow5DeadlineBudget5With the increase in processing power and computation tools, computation has become a third branch in scientific research along theory and experiment. Scientific worklfow aims to automate these complex analysis and facilitates collaborative research.Research overviewJust-in-time and adaptive scheduling heuristic Using spot and on-demand instancesAn intelligent bidding strategyMinimizes the execution cost Providing a robust scheduleSatisfying the deadline constraint6To the best of our knowledge, there has been no study in workflow scheduling algorithm for Clouds maximizing robustness, and minimizing makespan and cost at the same time. Also there are very few works which scheduleworkflow tasks on heterogeneous Cloud resources. This study tries to address these shortcomings.6BackgroundWorkflow is represented a DAGMakespan is the total elapsed timePricing modelsOn-DemandSpotCritical Path is the longest path from the start node to the exit node

Latest Time to On-Demand (LTO)It is the latest time the algorithm has to switch to on-demand instances to satisfy the deadline constraintDeadlineLTOStartSpot InstancesOn-DemandSystem Model

Runtime EstimationWe use Downeys analytical modelDowneys model requires:tasks average parallelism, A, coefficient of variance of parallelism, , task length the number of coresCirne et al model to generate A and Failure EstimatorEstimates the failure probability of a particular bid priceBased on spot priceThe history price of one month prior is consideredTotal time of the spot price history, HTAnd total out of bid time, OBTbidt is measured

Scheduling Algorithm

Scheduling Algorithm (Contd..)

Scheduling Algorithm (Contd..)

Two type of Scheduling AlgorithmsConservative: CP and LTO is estimated on the lowest cost instance. CP is the longest, hence less slack timeUses spot instances cautiously under relaxed deadlinesAggressive: CP and LTO is estimated on the highest cost instance. CP is smallest, hence more slack timeopt on-demand instances that are expensive under failuresBidding StrategyIntelligent Bidding StrategyCurrent spot price (pspot)On-demand price (pOD)Failure probability (FP) of the previous bid priceLTOCurrent time (CT)

Intelligent Bidding Strategy : dictates how much higher the bid value must be above the current spot price : determines how fast the bid value reaches the on-demand priceFP of the previous bid is used as a feedback to the current bid price

Intelligent Bidding Strategy

Other Bidding StrategiesOn-Demand Bidding Strategy : uses the on-demand price as the bid price.Naive Bidding Strategy: uses the current spot price as the bid price for the instanceSimulation SetupCloudSim was used for simulationLIGO workflow with 1000 tasks was consideredFor On-Demand 9 different VMs types wereconsideredFor Spot, 1 VM type was used Results : Comparison between algorithms

Mean execution cost of algorithms with varying deadline (with 95% confidence interval)Results : Comparison between bidding strategiesMean Execution Cost of bidding strategies with varying deadline (with 95% confidence interval)

Results : Task FailuresMean of task failures due to bidding strategies

Results : Checkpointing

ConclusionTwo scheduling heuristics that map workflow tasks onto spot and on-demand instance are presentedThey minimize the execution costThey are robust and fault-tolerant towards out-of-bid failures and performance variationsA bidding strategy that bids intelligently to minimize the cost is presentedDemonstrates the use of checkpointing, which offers cost savings up to 14%

Copyright The University of Melbourne 2009


Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.