DBTCP: A Transmission Mechanism of the Backup for Disaster Recovery in DCNs

Abstract

End-to-end communication has been widely used in all aspect of our daily lives, and are these applications are mostly rely on data center networks (DCNs), which are served as the core infrastructure for storing and processing a large number of data. Disaster backup, different from general data transmission, has some unique characteristics and requires low latency, high throughput, burst robustness, tight deadline, massive loads and easy deployment, where mass data in DCNs nodes need to be transferred as much as possible and the tight deadline should be meet after the alarm for a potential nature extreme phenomena is sound. This paper first introduces and summarizes the existing transmission mechanisms deployed for DCNs worldwide. Further, we explain the unique characteristics of disaster backup. Then our paper discusses the design idea and proposes the detailed design of the DCN backup TCP (DBTCP), through simulation and evaluation, the results of which displayed in the form of graphs. We are confirmed that the DBTCP designed is capable of making almost all of flows meet the tight deadline under high load and high burst. By carefully investigations implemented in both NS2 and Linux kernel, we notice that DBTCP nearly always outperforms previous presented mechanisms, like DCTCP, D^2TCP. In typical scenarios, DBTCP can reduce missing deadline rate and improve both throughout and robustness when flows burst, keeping the outstanding easy deployment feature of former mechanisms at the same time. The DBTCP proposed in this paper provides a reliable and effective solution, and has a certain reference value for future research.