IBM Research Leveraging Attention Scarcity to Improve the Overall User Experience of Cloud Services Marco Netto, Marcos Assunção and Silvia Bianchi © 2013 IBM Corporation
Background § Large volumes of data are produced every day § Increasing use of mobile devices * § Around 700 million smartphones shipped a year § 1. 5 billion mobile broadband subscriptions by end 2013 § Cloud-computing as a platform to support user services § Multiple devices and media multi-tasking § 53% of UK adults media multi-task while watching TV ** § Users expect other devices to augment content shown on TV *** * The State of Broadband 2013: Universilizing Broadband, Broadband Commission, Sep. 2013. ** The Communications Market Report: United Kingdom, Ofcom, Aug. 2013. *** Five insights into consumers’ online video viewing and buying habits, Accenture Point of View, No. 2 Jul. 2013. 2 © Copyright IBM Corporation 2013
Motivation § Applications, information and media are abundant, attention is scarce § Cloud is elastic, but resources are limited § Current devices can provide more honest signals on user behaviour and their profiles * § When do users actually consume the results of their requests? * http: //www. drawbrid. ge http: //www. flurry. com/ http: //www. velti. com 3 © Copyright IBM Corporation 2013
Illustrative Scenario and Research Question § User switches across multiple applications Game App ? ? Active Apps § Can these signals on service consumption be used to differentiate the processing of requests, optimise the use of Cloud resources, and deliver better user experience? 4 © Copyright IBM Corporation 2013
Problem Description § Consider two service requests, req 1 and req 2, made by two distinct users § User patience di = tci – rti where d < 0 means user is angry § User experience ue is modelled using principles from prospect theory, where d < 0 translates into worse ue than under |d| 5 © Copyright IBM Corporation 2013
§ Goals of the present work: § Minimising the distance between consumption time and response time § Investigate the impact on overall user experience § Outline: § Adaptive Qo. S architecture § Request prioritisation strategy § Experimental evaluation 6 © Copyright IBM Corporation 2013
Adaptive Qo. S Architecture Qo. S Setup Service User Device Cloud App 1 IT Infrastructure Service 1 Cloud App 2 Service 2 Local App Service n Sensors Qo. S Setup Assistant 7 Cloud Service Provider © Copyright IBM Corporation 2013
Qo. S Setup Service § Determines the user context § Monitors how users interact with a service § Patience-aware Prioritisation (Pa. P): 1. Obtains the expected time before consumption and average response time 2. Determines the user’s patience 3. Prioritises users who have less patience 4. To avoid starvation, a user’s patience decreases as her requests waits to be served 8 © Copyright IBM Corporation 2013
Experimental Setup § Built-in-house discrete event simulator § One service able to handle 100 requests at a time § Two categories of users, namely multi-task and single task § tc’s of single-task and multi-task users are 2 and 10 seconds respectively § Requests have fixed processing length § Evaluation under three workload scenarios, comparing Pa. P against a FCFS (standard) policy * * C. Cardonha, M. Assuncao, M. Netto, R. Cunha, C. Queiroz, Patience-aware Scheduling for Cloud Services: Freeing Users from the Chains of Boredom, ICSOC 2013. 9 © Copyright IBM Corporation 2013
Overall User Patience Single-task users Multi-task users 10 © Copyright IBM Corporation 2013
User Experience (Scenario 3) Pa. P strategy Standard strategy 11 © Copyright IBM Corporation 2013
User Experience (Scenario 2) Pa. P strategy Standard strategy 12 © Copyright IBM Corporation 2013
Final Remarks § Important to understand how users consume results of Cloud services and to optimise the utilisation or resources or enhance user experience § An architecture for leveraging more honest signals § A simple strategy to prioritise user requests * § Possible to improve user experience without heavily impact on users who multi-task § Proposed strategy is suitable to peak load conditions * 13 C. Cardonha, M. Assuncao, M. Netto, R. Cunha, C. Queiroz, Patienceaware Scheduling for Cloud Services: Freeing Users from the Chains of Boredom, ICSOC 2013. © Copyright IBM Corporation 2013
Thank you for your attention! 14 © Copyright IBM Corporation 2013
Скачать презентацию IBM Research Leveraging Attention Scarcity to Improve the
59b0d0f67a7dd42a66d3826a6ebd0ef5.ppt