Abstract:
The Internet of Things (IoT) and the Cloud Computing are the two emerging technologies which have seen a wide-scale adoption. Many IoT applications require cloud platform support for computing and storage purpose. Hence managing network resources and congestion control in a cloud data center network (DCN) environment have become a major concern. In a large scale DCN, there are several storage and network devices interconnected. Achieving high performance in an environment having low latency, high throughput and shallow buffered network devices, poses a great challenge for network administrators. In such situation many-to-one TCP communication pattern predominates, which often results in an individual link throughput collapse known as Incast. In this paper, we present a centralized approach, called as modified dynamic fair-share buffer policy (M-DyFaShBuP), which operates in a high bursty traffic environment. The software-defined network (SDN) controller plays an imperative role from policy designing till policy implementation. An equitable share for each server is calculated and distributed by the controller among all storage servers. The storage servers adjust their sending rates, according to the received policy. We incorporate topof- rack (TOR) and end-of-row (EOR) designs while handling Incast. M-DyFaShBuP successfully tackles the Incast scenario and achieves high throughput performance.
