PROJECT TITLE :

User-Centric Interference-Aware Load Balancing for Cloud-Deployed Applications

ABSTRACT:

Virtual machines that are hosted in cloud environments are more likely to experience performance issues as a result of the dynamic and unpredictable competition among colocated tenants for shared physical resources. Current provider-centric solutions, such as carefully co-scheduling of virtual machines (VMs) and/or VM migration, require a priori profiling of customer VMs, which is not feasible in public clouds. Examples of such solutions include: In addition, these solutions do not always take into account the SLO requirements of the user or the application bottlenecks. This paper introduces DIAL, an interference-aware load balancing framework that cloud users can directly employ without needing any assistance from the provider. DIAL can also be used by users in conjunction with existing load balancing solutions. The most important concept that underpins DIAL is the concept of inferring the demand for congested resources on physical hosts, which is normally kept secret from the users. Following this, estimates of the colocated load are used to dynamically shift load away from compromised VMs in a manner that does not violate the application's tail latency SLOs. We implement DIAL for web and online analytical processing applications, and show, via experimental results on OpenStack and AWS clouds, that DIAL can reduce tail latencies by as much as 70 percent compared to existing solutions. This is demonstrated by the fact that DIAL is able to reduce tail latencies by an order of magnitude.