Securing cloud resource consumption using machine learning

Abstract

Benefits like agility, convenience, easy to manage, flexibility and budget-friendly are newlinesome of the predominant reasons for consumers to rush to embrace public cloud services like newlineAmazon Web Services (AWS), Microsoft Azure, and others. This new technology is rapidly newlinegaining popularity as an attractive alternative to private clouds or on-premises physical servers. newlinePublic Cloud has become popular because of its utility modeling and a shift from capital newlineexpenses to operational expenses. More and more consumers are moving to the public cloud and newlinehosting their applications on cloud cost-effectively, to get the benefit of high availability and newlinescalability. But this comes with security tradeoff. Consumers still worry about cloud security as newlinethere are more security vulnerabilities to address. Cloud Utility modeling helps the consumer to newlinescale up or scale down the resources dynamically as required. Challenge relies on detecting that newlinethese resources are not maliciously consumed. Malicious resource consumption can lead to the newlinefinancial viability of the cloud consumer. Web robots cause a serious threat to cloud resources by newlinemaliciously consuming and exhausting them slowly. Malicious Cloud Bandwidth Consumption newline(MCBC) is one type of threat, where attackers consume the cloud bandwidth slowly and newlinecontinuously for an extended time causing the financial burden to the cloud consumer. The goal newlineof the MCBC attack is to impact the billing structure of cloud-hosted service by directly newlineaffecting the cost of service provisioning. Unlike Distributed Denial of Service (DDoS) which newlinecan be detected based on an increase in traffic volume, MCBC attacks may not be easily newlinedetected, since MCBC requests mimic the legitimate requests and most of the time it goes newlineundetected. Securing cloud resource consumption plays a major role in protecting the financial newlineviability of the consumer. This research work emphasizes on securing cloud web resource newlineconsumption by detailing about MCBC attack and proposes a methodology for building newlineclassifiers using Machine Learning (ML) to accurately detect malicious requests. Once the threat newlineis detected, a clear estimation of the financial burden incurred from MCBC is computed to help newlinethe consumer to take quick action on the threat. Experimental results apart from classifying newlineMCBC attacks also demonstrate that it is challenging for an attacker to carry out a successful newlineMCBC attack without the actual knowledge of web access logs. newline