Research Statement Faisal Nawab

Processing large quantities of data is becoming more ubiquitous and is the driving force behind the sustained growth and impact of Internet Services and Big Data analytics. The way data-intensive applications are deployed has been radically transformed by the cloud computing paradigm realized through massive-scale datacenters. However, datacenter-scale failures have occurred numerous times in the past and continue occurring many times annually due to various events such as power outages and natural disasters. These failures impact all services in a datacenter for extended periods of time and cause disruption to a large number of users, leading to losses in revenue and utility of Internet and Big Data applications. Also, deployment at a single datacenter gives rise to very high latency for user requests that are geographically distant. Recently, Big Data and large-scale systems have addressed datacenter-scale failures and high service response times by distributing applications and data globally across multiple datacenters, thus providing datacenter-scale fault-tolerance and data proximity to users. Methods based on asynchronous replication are widely used in practice to keep the replicas up-todate with each other. Asynchronous replication, however, is vulnerable to data losses and inconsistency when failures occur, which threaten the integrity of the applications. This also complicates application design and recovery for developers and system administrators. This is now broadly recognized as a serious drawback that is limiting the adoption of multi-datacenter deployments. This has led to a growing interest from both industry and academia in global-scale systems with stringent consistency guarantees.