IP2DC: Making Sense of Replica Selection Tools

Abstract

Cloud-based applications being developed for consumer electronics market (tablets, smart TVs) struggle to deliver the same level of responsiveness as standalone software, leading to user frustration and slow adoption. Often the network that separates user end-hosts from server back-end is to blame. To limit the impact of poor network performance on message delay, or lag, back-end logic and application data are deployed across geographically distributed servers and user requests are directed to the closest one [14]. Such nearby servers deliver content more quickly thanks to a faster expansion of TCP congestion window and more rapid retransmissions over low round-trip time (RTT) paths. Our early results, presented in this poster, show a high level of discrepancy between the available tools and motivate further measurement as well as the need to develop techniques for more accurate server replica selection.