CloudTalk: Enabling Distributed Application Optimisations in Public Clouds

Abstract

Clouds offer an opaque I/O API to their customers: details of the underlying resources (network topology, disk drives) or their current load are kept hidden. Tenants can profile the I/O performance in their VMs and optimise accordingly, but the side effect is increased load. Certain cloud providers try to discourage profiling by enforcing strict I/O isolation, at the cost of reduced utilisation in the average case. In this paper we challenge this status quo and propose CloudTalk, an API that allows tenants to communicate with the cloud provider and receive hints used to optimise their workloads. We have built a distributed implementation of CloudTalk that scales to hundreds of machines and provides significant performance benefits in many cases. Further, we have implemented changes to Hadoop and HDFS that use CloudTalk to decide which machines to use for task placement and replica selection. Our experiments in a local cluster and on Amazon EC2 show that CloudTalk helps improve performance by as much as two times for a wide range of scenarios.