Enabling Cost-effective Elasticity in Cloud Computing
撰稿：
摄影：
发布时间：2013年07月02日
时间：7月15日上午10：00—11：30
地点：计算所446会议室
Abstract:
As internet is becoming more fast, reliable and more ubiquitous, the pay-as-you-go model supported by existing Infrastructure-as-a-Service (IaaS) Cloud providers is appealing to most application owners (cloud consumers) to deploy and scale their applications in the Cloud, thus paying only for the amount of resources they use. Within this context, elasticity (on-demand scaling) of applications becomes one of the most important features of a cloud computing platform. This elasticity enables real-time acquisition/release of computing resources to meet application performance demands.
Traditionally, the application level elasticity addresses the question of how toscale the applications themselves up and downto meet their Quality of Service (QoS) requirements,while minimising the application owners’ operational cost. For such an issue, resource-price-aware criteria and workload-adaptive approaches should be developed to discover the real bottlenecks of the applicationsto bescaled. Also, lightweight resource scaling approachthat conducts the fine-grained application scaling at the hardware resource level itself (CPUs, memory, I/O, etc) in addition to VM-level scaling is discussed. Taking a multi-tier e-commerce website as an example to capture the typical behaviours of cloud applications, the effectiveness of both approaches are demonstrated.
Existing application level elasticity assumes that there is a one-off answer in its service—either it produces a result or it fails to produce a result. This assumption needs to be revisited when dealing with large-scale data under resource and time constraints(e.g., search an image or a webpage in Baidu within 5 seconds)—users are usually willing to accept approximate results produced using their available time budget in this case. Elasticity management at the algorithm level, coupled with the Pay-as-you-go cloud business models, give rise to twonew challenges about how we design programs and algorithms.The first is to reason about tradeoffs between the quality of output result, e.g. prediction accuracy for classification tasks, and available resource and time budgets. The second is organizing the computation of the algorithm to guarantee producing better quality of results as more budget is used.Using awidely applied service, namely collaborative filtering (CF) recommendation, in current E-commerce websites, weshow how both challenges are addressed in the framework of elastic algorithm.
Bio:
Rui Han isa final-year PhD Candidate at the Department of Computing, Imperial College London, UK.He received MSc with honorin 2010 from Tsinghua University, China. His research interests are cloud computing, cloud resource management and elastic algorithms for large-scale data analysis.In his current research he tries to combine the advantages of Pay-as-you-Go model supported by existing IaaS Cloud providers and data analysis algorithms in order to deliver elastic services in the Cloud.He has authored and co-authored over 15 publications in refereed journals and conferences, and he served in a series of academic conferences.
附件下载：
Enabling Cost-effective Elasticity in Cloud Computing
地点：计算所446会议室
Abstract:
As internet is becoming more fast, reliable and more ubiquitous, the pay-as-you-go model supported by existing Infrastructure-as-a-Service (IaaS) Cloud providers is appealing to most application owners (cloud consumers) to deploy and scale their applications in the Cloud, thus paying only for the amount of resources they use. Within this context, elasticity (on-demand scaling) of applications becomes one of the most important features of a cloud computing platform. This elasticity enables real-time acquisition/release of computing resources to meet application performance demands.
Traditionally, the application level elasticity addresses the question of how toscale the applications themselves up and downto meet their Quality of Service (QoS) requirements,while minimising the application owners’ operational cost. For such an issue, resource-price-aware criteria and workload-adaptive approaches should be developed to discover the real bottlenecks of the applicationsto bescaled. Also, lightweight resource scaling approachthat conducts the fine-grained application scaling at the hardware resource level itself (CPUs, memory, I/O, etc) in addition to VM-level scaling is discussed. Taking a multi-tier e-commerce website as an example to capture the typical behaviours of cloud applications, the effectiveness of both approaches are demonstrated.
Existing application level elasticity assumes that there is a one-off answer in its service—either it produces a result or it fails to produce a result. This assumption needs to be revisited when dealing with large-scale data under resource and time constraints(e.g., search an image or a webpage in Baidu within 5 seconds)—users are usually willing to accept approximate results produced using their available time budget in this case. Elasticity management at the algorithm level, coupled with the Pay-as-you-go cloud business models, give rise to twonew challenges about how we design programs and algorithms.The first is to reason about tradeoffs between the quality of output result, e.g. prediction accuracy for classification tasks, and available resource and time budgets. The second is organizing the computation of the algorithm to guarantee producing better quality of results as more budget is used.Using awidely applied service, namely collaborative filtering (CF) recommendation, in current E-commerce websites, weshow how both challenges are addressed in the framework of elastic algorithm.
Bio:
Rui Han isa final-year PhD Candidate at the Department of Computing, Imperial College London, UK.He received MSc with honorin 2010 from Tsinghua University, China. His research interests are cloud computing, cloud resource management and elastic algorithms for large-scale data analysis.In his current research he tries to combine the advantages of Pay-as-you-Go model supported by existing IaaS Cloud providers and data analysis algorithms in order to deliver elastic services in the Cloud.He has authored and co-authored over 15 publications in refereed journals and conferences, and he served in a series of academic conferences.