Control-theoretical load-balancing for cloud applications with brownout

Abstract

Abstract-Cloud applications are often subject to unexpected events like flash crowds and hardware failures. Without a predictable behaviour, users may abandon an unresponsive application. This problem has been partially solved on two separate fronts: first, by adding a self-adaptive feature called brownout inside cloud applications to bound response times by modulating user experience, and, second, by introducing replicas -copies of the applications having the same functionalities -for redundancy and adding a load-balancer to direct incoming traffic.However, existing load-balancing strategies interfere with brownout self-adaptivity. Load-balancers are often based on response times, that are already controlled by the self-adaptive features of the application, hence they are not a good indicator of how well a replica is performing.In this paper, we present novel load-balancing strategies, specifically designed to support brownout applications. They base their decision not on response time, but on user experience degradation. We implemented our strategies in a selfadaptive application simulator, together with some state-of-theart solutions. Results obtained in multiple scenarios show that the proposed strategies bring significant improvements when compared to the state-of-the-art ones.

BrownoutCC: Cascaded Control for Bounding the Response Times of Cloud Applications

“…EVALUATION This section presents our results. We validate our control strategy using the open source Python-based brownout simulator 2 , built to mimic the behavior of cloud applications [10] and described in the following Section IV-A.…”

“…Computing resource allocation can easily be cast into a control problem, where a controller decides the amount of resource to allocate to different entities based on desired and measurable performance metrics [1,17,21,25]. Recently, the cloud computing domain has emerged as an interesting application domain for control-theoretical principles and techniques [4,7,10,15,22].…”

Improving Cloud Service Resilience Using Brownout-Aware Load-Balancing

“…Preliminary simulation results [13] compared different loadbalancing algorithms for this architecture, such as round-robin, fastest replica first, random and two random choices. The main result of this comparison is that load-balancing algorithms that are based on measurements of the response times of the single replicas are not suited to be used with brownout-compliant services, since the replica controllers already keep the response times close to the setpoint.…”

“…The main result of this comparison is that load-balancing algorithms that are based on measurements of the response times of the single replicas are not suited to be used with brownout-compliant services, since the replica controllers already keep the response times close to the setpoint. The only existing algorithm that proved to work adequately with brownout-compliant services is Shortest Queue First (SQF) [10], [13]. It works by tracking the number of queued requests q i on each replica and directing the next request to the replica with the lowest q i .…”

“…However, to date, only weight-based algorithms were considered, where each replica gets a fraction of the incoming traffic proportional to a dynamic weight. A controller periodically adjusts the weights based on the dimmer values of each replica [13]. Results suggested that deciding periodically gives good results in steady-state, however, the resulting service is not reactive enough to sudden capacity changes, as would be the case when a replica fails.…”