ExPERT: Pareto-efficient task replication on grids and a cloud Orna - PowerPoint PPT Presentation

ExPERT: Pareto-efficient task replication on grids and a cloud Orna Agmon Ben-Yehuda 1 Assaf Schuster 1 Artyom Sharov 1 Mark Silberstein 1 Alexandru Iosup 2 1 Department of Computer Science Technion — Israel Institute of Technology 2 Faculty of Engineering, Mathematics and Computer Science (EWI) TU Delft IPDPS, May 2012 Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 1/33

The Shared Resource Game — Players and Goals User goals − minimize: Owner goals − minimize: *Makespan *Operational costs (energy) *Cost *Effective load Policy Enforcing QoS Costs Resource USER Resource OWNER Workload Paying for resources or QoS Strategy (Declarations, Resource Usage) Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 2/33

The Unreliable Shared Resource Game User goals − minimize: Owner goals − minimize: *Makespan *operational costs (energy) *Cost *effective load Policy Enforcing QoS Costs Unreliable (Preemption) Slow/Costly Grid USER Grid OWNER Reliable Bag of Async Tasks Alternative Paying for Credentials An environment of uncertainty: Will the task fail on the unreliable resource? Which system to use? Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 3/33

In the Beginning... Failure/ D Timeout Unreliable Unreliable queue Pool Success # machines < # unfinished tasks . D - instance deadline. No replication (replication is inefficient). Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 4/33

Using the Same Strategy After the Tail Starts # machines > # unfinished tasks 600 remaining tasks Remaining tasks Number of Tail phase start time ( T tail ) 400 200 Tail Phase Throughput Phase 0 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 Time [s] The tail is wagging the dog... Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 5/33

Replication - the User’s Bank of NTDM r Strategies Failure/ First N Tail Instances T D Timeout Unreliable Unreliable queue Pool Success T Instance N+1 in Tail Reliable Reliable Queue Pool Success D - instance deadline, T - replication time Reliable machine used to ensure task completion N tail instances at most on unreliable resources M r - max ratio of reliable to unreliable resources Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 6/33

Replication - the User’s Bank of NTDM r Strategies Failure/ First N Tail Instances T D Timeout Unreliable Unreliable queue Pool Success T Instance N+1 in Tail Reliable Reliable Queue Pool Success D - instance deadline, T - replication time Reliable machine used to ensure task completion N tail instances at most on unreliable resources M r - max ratio of reliable to unreliable resources Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 6/33

Replication - the User’s Bank of NTDM r Strategies Example: Number of unreliable instances N = 3 ������ ������ ������� ������� UNRELIABLE1 ������ ������ ������� ������� ������� ������� UNRELIABLE2 ������� ������� �� �� �� �� �� �� � � RELIABLE � � � � 0 T D 2T 3T 3T+Tr Time Replication wastes work! Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 7/33

The User’s Problem: Optimization of... The user cares about multi-objective optimization: � Cost � - Mean cost task or tail − cost tail − task � MS � - Mean makespan or tail makespan. Each user may have her own objective, pending on those values: Below minimal makespan: � MS � < Const As fast as possible: min � MS � Below max budget: � Cost � < Const As cheap as possible: min � Cost � Best price for the goods: min � Cost �� MS � Any other function of means: � Cost � , � MS � . . . Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 8/33

The Feedback Loop Directly Indirectly Users who Heavy Users are Owners Credential Billed Users Caring Users Costs Trial & Trial & Error Error USER OWNER Strategy Users who do not optimize well behave irrationally and are hard to predict. Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 9/33

The Feedback Loop — Our Contribution Directly Indirectly Users who Heavy Users are Owners Credential Billed Users Caring Users Costs Trial & Optimize Error USER Strategy OWNER Rational users can optimize general utility function. Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 10/33

The Feedback Loop - Lookout Directly Indirectly Users who Heavy Users are Owners Credential Billed Users Caring Users Costs Optimize Optimize USER Strategy OWNER Towards the final goal of manipulating users to save energy Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 11/33

Solution Concept ExPERT 3 Pareto 1 Decision 2 Frontier Making Statistical Generation Characterization 4 User Unreliable Pool Reliable Pool Scheduler 5 BoT Execution Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 12/33

Solution Concept - Step 1 Get user additional data (costs, reliable pool times). ExPERT 3 Pareto 1 Decision 2 Frontier Making Statistical Generation Characterization 4 User Unreliable Pool Reliable Pool Scheduler 5 BoT Execution Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 13/33

Solution Concept - Step 2 Get unreliable resource statistics (trace analysis). ExPERT 3 Pareto 1 Decision 2 Frontier Making Statistical Generation Characterization 4 User Unreliable Pool Reliable Pool Scheduler 5 BoT Execution Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 14/33

Solution Concept - Step 3 Compute a Pareto frontier for � Cost � , � MS � . ExPERT 3 Pareto 1 Decision 2 Frontier Making Statistical Generation Characterization 4 User Unreliable Pool Reliable Pool Scheduler 5 BoT Execution Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 15/33

Solution Concept - Step 3 Estimate � Cost � , � MS � for each strategy in the search space: For every working point, the ExPERT Estimator computes several random realizations on the basis of the statistic characterization. The average maksespan and cost over these realizations are used as the expectation values � Cost � , � MS � . ExPERT 3 Pareto 1 Decision 2 Frontier Making Statistical Generation Characterization 4 User Unreliable Pool Reliable Pool Scheduler 5 BoT Execution Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 16/33

Solution Concept - Step 3 Estimate � Cost � , � MS � for each strategy in the search space. Filter out dominated strategies. Keep frontier composed of non-dominated strategies. Cost Makespan Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 17/33

Solution Concept - Step 3 Estimate � Cost � , � MS � for each strategy in the search space. Filter out dominated strategies. Keep frontier composed of non-dominated strategies. Dominated Area Cost Makespan Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 17/33

Solution Concept - Step 3 Estimate � Cost � , � MS � for each strategy in the search space. Filter out dominated strategies. Keep frontier composed of non-dominated strategies. Dominated Dominated Area Strategy S 3 Cost Non−dominated Strategy S 2 Non−dominated StrategyS 1 Makespan Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 17/33

Solution Concept - Step 3 Estimate � Cost � , � MS � for each strategy in the search space. Filter out dominated strategies. Keep frontier composed of non-dominated strategies. Dominated Area Cost Non−dominated Strategy S 2 Non−dominated StrategyS 1 Makespan Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 17/33

Solution Concept - Step 3 Estimate � Cost � , � MS � for each strategy in the search space. Filter out dominated strategies. Keep frontier composed of non-dominated strategies. Cost Non−dominated Strategy S 2 Non−dominated Pareto Frontier StrategyS 1 Makespan Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 17/33

Solution Concept - Step 4 Choose optimal strategy according to user utility (by expectation value ). ExPERT 3 Pareto 1 Decision 2 Frontier Making Statistical Generation Characterization 4 User Unreliable Pool Reliable Pool Scheduler 5 BoT Execution Get N , T , D , M r params for the desired strategy. Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 18/33

Solution Concept - Step 5 Apply strategy: Feed N , T , D , M r params as input to the user scheduler and deploy tasks on the resource pools. ExPERT 3 Pareto 1 Decision 2 Frontier Making Statistical Generation Characterization 4 User Unreliable Pool Reliable Pool Scheduler 5 BoT Execution Agmon Ben-Yehuda, Schuster, Sharov, Silberstein, Iosup ExPERT 19/33