Web application providers have been migrating their applications to cloud data centers, attracted by the emerging cloud computing paradigm. One of the appealing features of the cloud is elasticity. It allows cloud users to acquire or release computing resources on-demand, which enables web application providers to automatically scale the resources provisioned to their applications without human intervention under a dynamic workload to minimize resource cost while satisfying Quality of Service (QoS) requirements. In this paper, we comprehensively analyze the challenges that remain in auto-scaling web applications in clouds and review the developments in this field. We present a taxonomy of auto-scalers according to the identified challenges and key properties. We analyze the surveyed works and map them to the taxonomy to identify the weaknesses in this field. Moreover, based on the analysis, we propose new future directions that can be explored in this area.

Cloud providers sell their idle capacity on markets through an auction-like mechanism to increase their return on investment. The instances sold in this way are called spot instances. In spite that spot instances are usually 90% cheaper than on-demand instances, they can be terminated by provider when their bidding prices are lower than market prices. Thus, they are largely used to provision fault-tolerant applications only. In this paper, we explore how to utilize spot instances to provision web applications, which are usually considered availability-critical. The idea is to take advantage of differences in price among various types of spot instances to reach both high availability and significant cost saving. We first propose a fault-tolerant model for web applications provisioned by spot instances. Based on that, we devise novel cost-efficient auto-scaling polices that comply with the defined fault-tolerant semantics for hourly billed cloud markets. We implemented the proposed model and policies both on a simulation testbed for repeatable validation and Amazon EC2. The experiments on the simulation testbed and EC2 show that the proposed approach can greatly reduce resource cost and still achieve satisfactory Quality of Service (QoS) in terms of response time and availability.

Abstract-Cloud computing is an utility computing paradigm that allows users to flexibly acquire virtualized computing resources in a pay-as-you-go model. To realize the benefits of using cloud, users need to first select the suitable cloud services that can satisfy their applications' functional and non-functional requirements. However, this is a difficult task due to large number of available services, users' unclear requirements, and performance variations in cloud. In this paper, we propose a system that evaluates trust of clouds according to users' fuzzy Quality of Service (QoS) requirements and services' dynamic performances to facilitate service selection. We demonstrate the effectiveness and efficiency of our system through simulations and case studies.

No abstract

Managed by an auto-scaler in the clouds, applications may still be overloaded by sudden flash crowds or resource failures as the auto-scaler takes time to make scaling decisions and provision resources. With more cloud providers building geographically dispersed data centers, applications are commonly deployed in multiple data centers to better serve customers worldwide. In this case, instead of sufficiently over-provisioning each data center to prepare for occasional overloads, it is more cost-efficient to over-provision each data center a small amount of capacity and to balance the extra load among them when resources in any data center are suddenly saturated.In this paper, we present a decentralized system that timely detects short-term overload situations and autonomously handles them using geographical load balancing and admission control to minimize the resulted performance degradation. Our approach also includes a new algorithm that optimally distributes the excessive load to remote data centers causing minimum increase of overall response times. We developed a prototype and evaluated it on Amazon Web Services.The results show that our approach is able to maintain acceptable quality of service while greatly increase the number of requests served during overloading periods.