摘要
In the past few years, Big Data analytics have changed the way computing services and resources are being used. New users are getting into the cloud services provided by data centers on a daily basis, and the number of applications hosted on the clouds is rapidly increasing. However, existing data center architectures cannot cope with the black-box nature of the cloud nor with Big Data analytics. Current data center networks using electrical packet switches have many disadvantages. Electrical links consume too much power to handle the increased bandwidth demanded by emerging applications. They are often not suitable for providing high bandwidth at low latency and have bottleneck issues. A promising solution involves adding optical interconnections to the infrastructure in data centers offering high throughput, low latency, and incredibly low energy consumption compared to the current data center networks based on commodity switches. In this work, we propose a hybrid electrical and optical networking architecture for data centers hosting Cloud Computing and Big Data applications. Furthermore, we analyze and compare application behaviors with current and proposed hybrid architectures. The results show that hybrid architecture provides a viable solution to data center issues and easily outperforms existing data center architectures. The results reflect that hybrid architecture reduces the delay by nearly 39% and decreases the cost of cooling systems from 49.57% to 27% of the total costs, while reducing the overall long-term costs of data centers.