Comprehensive link sharing avoidance and switch aggregation for software-defined data center networks

Highlights

• We formulate the network energy consumption minimization problem under the constraints of traffic conservation, link capacity and state relation.

• We propose the link sharing avoidance algorithm from time dimension as well as the switch aggregation algorithm from power dimension to achieve energy saving.

• We propose a heuristic integrated time and power (ITP) algorithm to further reduce energy consumption.

• Experiment results show that our proposed algorithm has more energy saving and lower network delay for different network topologies, network scales, flow sizes and flow arrival rates.

Abstract

An effective way to reduce network energy consumption of data center networks (DCNs) is to activate network elements as few as possible, complete transmission in as short a time as possible, and set unnecessary network elements to sleep mode. At present, most existing energy saving works considered the network energy saving from the dimension of time or power separately. However, in fact these two dimensions can interact with each other, i.e., reducing the network delay may lead to the increase of network energy consumption, and vice versa. In this paper, two dimensions of time and power are comprehensively studied in the Minimum Network Energy Consumption (MNEC) problem. First of all, we formulate the MNEC problem by considering both time and power, and prove that it is a NP-hard problem. Furthermore, we propose a heuristic Integrated Time and Power (ITP) algorithm, which combines the link sharing avoidance algorithm to reduce the network delay from the dimension of time as well as the switch aggregation algorithm to reduce the energy consumption from the power dimension. Finally, the performance of ITP algorithm is evaluated under different network topology, network size, traffic size and flow number under the network environment based on Mininet and Ryu controller. Experimental results show that the ITP algorithm outperforms the existing network energy saving algorithm in terms of energy consumption.

Introduction

In recent years, with the development of cloud computing and video services, the scale and energy consumption of data centers have increased dramatically. In 2013, the energy consumption of data centers in the United States reached 91 billion kilowatt-hours, which was close to that of the Three Gorges hydropower station, the largest hydropower station in the world, with an annual capacity of 98 billion kilowatt-hours. In addition, it is expected that by 2020 the energy consumption will reach 140 billion kilowatt-hours [1]. Due to the rapid increase of energy consumption in data centers, energy saving has become one of the major bottlenecks of design data centers. In the meanwhile, the network energy consumption dominates the whole energy consumption in the data center, up to 15% [2]. Therefore, how to reduce data center network energy has become a hot topic in industry and academia.

In order to ensure that the data center has better performance, researchers have proposed numerous the network structures with redundant links, such as Fat-Tree [3], Bcube [4], FiConn [5]. Although redundant links and switches in network topology designed for meeting peak traffic requests can improve the performance of network bandwidth, they also lead to a sharp increase in network energy consumption since peak traffic is not regular at most of the time in the network. This means that running unnecessary links and switches will lead to the waste of network energy. An effective way to save energy is to activate as few switches and links as possible, finish transfers in as short time as possible, and sleep unnecessary network elements. Most existing works focus on local energy saving by considering how to make the switch component sleep for a long time when the switch is idle [6], [7], [8], or lets the switch adjust the sending rate according to network load [9]. In fact, switches can only obtain information from the surrounding switches and hosts, and cannot optimize the network energy globally in traditional networks.

Software Defined Networking (SDN) technology with separated control plane and data plane is emerged as a new network architecture. The SDN controller can obtain the global network information and control the data transmission in the whole network. Owing to the global network view, operators can collect topology information and forwarding state of the entire network. Meanwhile, the network flow can be adjusted dynamically, and more reasonable and energy-saving path for the flow is planned. Numerous network energy conservation works [10], [11], [12], [13], [14], [15] studied the network energy conservation from the global view of the network. These research works can be divided into two categories. From the power dimension, part of the research works [10], [11], [12], [13], [14] aggregated traffic into the smallest set of network elements, and let the idle network elements be sleep mode to reduce energy consumption. However, their traffic aggregation allows link sharing, which creates more network bottlenecks and reduces link utilization. From the dimension of time, part of the research works [15], [16] employed flow scheduling method to reduce the generation of network bottlenecks and network latency, so as to achieve the purpose of energy saving. However, these research works only plan a shortest path for the flow from idle links, rather than finding the path with minimum power based on the network state.

Different from the previous works of saving energy from time or power dimension separately, in this paper, we integrate two dimensions of power and time to find a path set for all traffic requests, activate links and switches in the path set, and configure unnecessary network components to sleep mode. From the time dimension, we propose a link sharing avoidance algorithm to improve the link utilization, avoid the network bottleneck and reduce the network transmission delay by making the flow occupy the link separately, so as to reduce the network energy consumption. In this algorithm, it is worth noting that if the given flow cannot find a path to avoid the shared link, then the path with minimum energy consumption for the shared link is planned for the flow. From the dimension of power, we propose a switch aggregation algorithm that improves the utilization of switches and reduces energy consumption without link sharing. By combining the link sharing avoidance algorithm with the switch aggregation algorithm, furthermore, we propose a heuristic Integrated Time and Power (ITP) algorithm to transform the path search problem of minimum network energy consumption into the minimum weight path planning problem. The link sharing avoidance algorithm and the switch aggregation algorithm jointly determine the path selection by setting the appropriate path weight. The weight of a path reflects the energy consumed in the path, that is, the smaller the weight, the less energy consumption. As a result, the time complexity of the minimum energy path planning problem is reduced to the time complexity of the minimum weight path search problem, so that the algorithm has a high real-time performance and can be deployed in large-scale networks.

The main contributions of this paper are summarized as follows:

• We formulate the Minimum Network Energy Consumption (MNEC) problem under the constraints of flow conservation, link capacity and state relation. Then, we prove that it is a NP-hard problem and transform it into the minimum weighted path planning problem in the weighted graph.

• We propose the link sharing avoidance algorithm from time dimension as well as the switch aggregation algorithm from power dimension to achieve energy savings. In particular, we propose a heuristic Integrated Time and Power (ITP) algorithm to further reduce energy consumption.

• We evaluate our algorithm in the experimental environment built by Mininet [17] and Ryu controller [18]. Experimental results show that our proposed algorithm has more energy savings under different network topologies, network scales, flow sizes and the number of flows.

The reset of the paper is organized as follows. In Section 2, we introduce the related work of network energy saving. In Section 3, we establish the MNCP model and proved that it is a NP-hard problem. Then, we propose the ITP algorithm that combines two dimensions of time and power to save network energy consumption in Section 4. In Section 5, the performance of our algorithm is tested through extensive simulation. Section 6 concludes this paper.