Yuya Tarutani

Virtual network reconfiguration for reducing energy consumption in optical data centers

Energy consumption by data centers has become a serious problem, and measures for its reduction should be developed. Such measures should address not only the energy consumption of the servers but also that of the network itself, because the latter is responsible for a substantial portion of the total energy consumption. One approach to reducing the energy consumption of the network within a data center is to use optical circuit switches (OCSs) at the core of the data center, where electronic switches are connected to the OCSs. In such a network, a virtual network can be configured by setting the OCSs to connect different ports of the electronic packet switches. Thus, the energy consumption of the network can be reduced by configuring the virtual network to minimize the number of ports required by the electronic packet switches and powering down any unused ports. In this paper, we propose a method called virtual network reconfiguration for data center networks (VNR-DCN) that immediately reconfigures the virtual network so as to reduce the energy consumption under the constraints on the bandwidth and delay between servers in data center networks based on optical communication paths. In VNR-DCN, we configure the virtual network to satisfy the requirements by setting the parameters of the topology, called generalized flattened butterfly, instead of solving an optimization problem. In the evaluation, we show that a virtual network configured by VNR-DCN requires a small number of active ports. In addition, we show the impact of virtual network configuration on energy consumption.

A virtual network to achieve low energy consumption in optical large-scale datacenter

A data center network should provide communication with sufficiently large bandwidth and small delay between all servers. On the other hand, energy consumption of the data center network should be minimized. To satisfy all of the above requirements, in this paper, we introduce the virtual network configured over the data center network constructed of the optical cross connects (OXCs) and the electronic switches. We design the virtual network topology (VNT) so as to achieve sufficiently large bandwidth and small delay with small energy consumption. To calculate the suitable VNT in a short period, we propose the topology called Generalized Flattened Butterfly and a method to set the parameters so as to suit the current condition. In our evaluation, we clarify that our method achieves the sufficient bandwidth and the target maximum number of hops between top-of rack(ToR) switches with small energy consumption.

Optical-Layer Traffic Engineering With Link Load Estimation for Large-Scale Optical Networks

Traffic information is required to perform optical-layer traffic engineering (TE). However, as the number of nodes in optical networks increases, the overhead for collecting the traffic volume information becomes large. In this paper, we develop a method that reduces the overhead for collecting traffic volume information by selecting a subset of nodes and by only collecting the traffic volume information from the selected nodes. Then, we estimate the traffic volume using the information gathered from the selected nodes. According to the simulation results, we clarify that our method can accurately identify the congested links in real ISP topologies, where the number of traffic demands passing through some links is large; however, the estimation errors of our method become large when the number of traffic demands passing each link is small. Furthermore, optical-layer TE can sufficiently mitigate congestion by using the traffic volume estimated by our method from the information on 50% of all nodes in the case of the Japan topology and 30% of all nodes in the case of the AT&T topology.

Temperature Distribution Prediction in Data Centers for Decreasing Power Consumption by Machine Learning

To decrease the power consumption of data centers, coordinated control of air conditioners and task assignment on servers is crucial. It takes tens of minutes for changes of operational parameters of air conditioners including outlet air temperature and volume to be actually reflected in the temperature distribution in the whole data center. Proactive control of the air conditioners is therefore required according to the predicted temperature distribution, which is highly dependent on the task assignment on the servers. In this paper, we apply a machine learning technique for predicting the temperature distribution in a data center. The temperature predictor employs regression models for describing the temperature distribution as it is predicted to be several minutes in the future, with the model parameters trained using operational data monitored at the target data center. We evaluated the performance of the temperature predictor for an experimental data center, in terms of the accuracy of the regression models and the calculation times for training and prediction. The temperature distribution was predicted with an accuracy of 0.095°C. The calculation times for training and prediction were around 1,000 seconds and 10 seconds, respectively. Furthermore, the power consumption of air conditioners was decreased by roughly 30% through proactive control based on the predicting temperature distribution.

Reducing Power Consumption in Data Center by Predicting Temperature Distribution and Air Conditioner Efficiency with Machine Learning

To reduce the power consumption in data centers, the coordinated control of the air conditioner and the servers is required. It takes tens of minutes for changes of operational parameters of air conditioners including outlet air temperature and volume to be reflected in the temperature distribution in the whole data center. So, the proactive control of the air conditioners is required according to the prediction temperature distribution corresponding to the load on the servers. In this paper, the temperature distribution and the power efficiency of air conditioner were predicted by using a machine-learning technique, and also we propose a method to follow-up proactive control of the air conditioner under the predicted optimum condition. Consequently, by the follow-up proactive control of the air conditioner and the load of servers, power consumption reduction of 30% at maximum was demonstrated.

IEEE1888 over WebSocket for Communicating across a Network Boundary

Some communications between devices are limited by network boundaries for ensuring security. In these cases, IEEE1888 does not provide inter-device communication. We thus propose IEEE1888 over WebSocket for communication between devices across network boundaries. In IEEE1888 over WebSocket, two devices are added to an IEEE1888 network for communication across network boundaries, so devices using traditional IEEE1888 need not to be changed. We implemented an experimental system of IEEE1888 overWebSocket. The results shows that IEEE1888 over WebSocket enables communication across network boundaries.

A network model for prediction of temperature distribution in data centers

We propose a novel network model for real-time prediction of temperature distribution in a data center so as to allow energy-efficient task assignment and facility management. We model various physical relationships in the data center as a network, including heat movements caused by airflow and heat generation by servers. Since changes in temperature distribution depend on physical properties of the data center such as equipment locations and server types, model parameters (connection weights in the network) that characterize relationship of nodes are determined by a machine learning technique using actual data center operation data. The proposed method provides prediction results in a shorter time than traditional methods such as model based on computational fluid dynamics and potential flow model, while maintaining prediction accuracy. We evaluate the performance of the proposed model through comparison with actual data from our experimental data center. The evaluation indicates that the proposed model can predict 10-minute future temperature distributions in 60 places in 3.3 ms, with a root mean square error of 0.49 degrees.

Large-scale ASP-based HEMS utilizing interactive web technologies

The concept of a Home Energy Management System (HEMS) as a power-saving technology for home appliances has received much attention. Most of current HEMSs based on the Smart Energy Profile (SEP), KNX, or ECHONET standards are implemented as closed systems within a Home Area Network (HAN); therefore, cost is a serious problem in the large-scale deployment of HEMSs. One possible way to reduce the overhead is to implement a HEMS as a cloud-based Application Service Provider (ASP) service. In such setup, a HEMS server is located in the cloud, and it monitors and controls home appliance devices over the Internet. We propose here a novel HEMS architecture using a cloud-based ASP service. We describe an overall architecture and some potential interactive Web communication technologies that realize a HEMS over the cloud-based ASP services. We also present a mathematical analysis and experimental results for estimating protocol overheads and server loads. These show that the WebSocket protocol is the most suitable protocol, allowing accommodation of twice as many households as other protocols. Finally, we describe our prototype implementation of the proposed architecture with the WebSocket protocol.