CANDIS: Heterogenous Mobile Cloud Framework and Energy Cost-Aware - PowerPoint PPT Presentation

CANDIS: Heterogenous Mobile Cloud Framework and Energy Cost-Aware Scheduling Sebastian Schildt, Felix B¨ usching, Enrico J¨ orns, Lars Wolf PhoneCom 2013

The Story So Far In Phonecom 2012 we presented a position paper DroidCluster: Towards Smartphone Cluster Computing The Streets are Paved with Potential Computer Clusters Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 2

CANDIS Architecture Developer Clients compile dex jar task.dex task.jar parameters results jar cdb Server control.jar task.cdb new dex Control ( ); task.dex cdb results.cdb Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 3

Scalability 100 XSLT Transformation 80 Execution Time [%] 60 40 20 0 1 2 3 4 5 Number of "Nexus 7" Devices Repeated XSLT transformation simulates a realistic workload Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 4

Adapting to Heterogenous Device Clouds Simple: Divide tasks into equal chunks Benchmarking: Divide tasks based on the results of a small test task Simple Scheduler Benchmarking Scheduler 100 90 80 70 t busy [%] 60 50 40 30 20 10 0 Nexus 7 U8860 LT26i Nexus One GT-I8160 Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 5

Electricity Price Variations 2012 Electricity is traded on a stock-like market Idea: Schedule computation, when electricity is cheap. Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 6

Scenario and Scheduler A task needs 72h of computation on available resources Every month the task needs to be finished before the 15th Simple scheduling strategy: Scheduler uses devices for computation when electricity price is below a threshold If a deadline miss can not be prevented otherwise, electricity will be bought at any price Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 7

Cost Savings Electricity will only be bought if it is less than 50 EUR/MWh Beat the average price by 21 % Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 8

Savings with more Aggressive Thresholds Threshold: 50 EUR/MWh Threshold: 30 EUR/MWh Beat average by 21 % Beat average by 33 % Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 9

Energy-price aware scheduling Highly fluctuating electricity prices Will continue with the shift to renewables unless somebody commercializes cheap fusion → in 50 years Load-shifting already well researched, but not applied on a large scale Nothing can load-shift easier than IT equipment Under-Utilization not a large cost factor for mobile devices: Leaving resources – which are there anyway – unused doesn’t hurt Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 10

Using Mobile Devices for Serious computation We are (still) convinced this is going to happen This will be the/a next big thing Probably there is money in it somewhere Can already see the first steps towards mainstream Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 11

Conclusions Mobile Device Clouds Will happen. Soon. Energy Cost-Aware Scheduling Makes a lot of sense. Candis code to play https://github.com/ejoerns/candis 谢 谢 谢谢 谢 谢 ! 有 有 有 问 问 问 题 题 题 吗 吗 吗 ？ ？ ？ Sebastian Schildt | CANDIS: Mobile Cloud Framework and Cost-Aware Scheduling | 12