Energy Efficient Scheduling of Cloud Application Components with Brownout

TL;DR

This paper proposes a brownout-based method for energy-efficient cloud application scheduling that selectively deactivates optional components to reduce energy use while managing overloads.

Contribution

It introduces a novel combined brownout approach that enhances energy savings and overload management in cloud data centers, especially for microservices.

Findings

Over 20% energy savings achieved.

Trade-offs identified between energy savings and user discounts.

Effective for self-contained microservices.

Abstract

It is common for cloud data centers meeting unexpected loads like request bursts, which may lead to overloaded situation and performance degradation. Dynamic Voltage Frequency Scaling and VM consolidation have been proved effective to manage overloads. However, they cannot function when the whole data center is overloaded. Brownout provides a promising direction to avoid overloads through configuring applications to temporarily degrade user experience. Additionally, brownout can also be applied to reduce data center energy consumption. As a complementary option for Dynamic Voltage Frequency Scaling and VM consolidation, our combined brownout approach reduces energy consumption through selectively and dynamically deactivating application optional components, which can also be applied to self-contained microservices. The results show that our approach can save more than 20% energy consumption and there are trade-offs between energy saving and discount offered to users.