Dependability where the mobile world meets the enterprise world
As we move toward increasingly larger scales of computing, complexity of systems and networks has increased manifold leading to massive failures of cloud providers (Amazon Cloudfront, November 2014) and geographically localized outages of cellular services (T-Mobile, June 2014). In this dissertation, we investigate the dependability aspects of two of the most prevalent computing platforms today, namely, smartphones and cloud computing. These two seemingly disparate platforms are part of a cohesive story—they interact to provide end-to-end services which are increasingly being delivered over mobile platforms, examples being iCloud, Google Drive and their smartphone counterparts iPhone and Android. In one of the early work on characterizing failures in dominant mobile OSes, we analyzed bug repositories of Android and Symbian and found similarities in their failure modes [ISSRE2010]. We also presented a classification of root causes and quantified the impact of ease of customizing the smartphones on system reliability. Our evaluation of Inter-Component Communication in Android [DSN2012] show an alarming number of exception handling errors where a phone may be crashed by passing it malformed component invocation messages, even from unprivileged applications. In this work, we also suggest language extensions that can mitigate these problems. Mobile applications today are increasingly being used to interact with enterprise-class web services commonly hosted in virtualized environments. Virutalization suffers from the problem of imperfect performance isolation where contention for low-level hardware resources can impact application performance. Through a set of rigorous experiments in a private cloud testbed and in EC2, we show that interference induced performance degradation is a reality. Our experiments have also shown that optimal configuration settings for web servers change during such phases of interference. Based on this observation, we design and implement the IC 2 engine which can mitigate effects of interference by reconfiguring web server parameters [MW2014]. We further improve IC 2 by incorporating it into a two-level configuration engine, named ICE, for managing web server clusters [ICAC2015]. Our evaluations show that, compared to an interference agnostic configuration, IC 2 can improve response time of web servers by upto 40%, while ICE can improve response time by up to 94% during phases of interference.
Bagchi, Purdue University.
Computer Engineering|Information Technology|Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our