Methods for Predicting Behavior of Elephant Flows in Data Center Networks

Abstract

Several Traffic Engineering (TE) techniques based on SDN (Software-defined networking) proposed to resolve flow competitions for network resources. However, there is no comprehensive study on the probability distribution of their throughput. Moreover, there is no study on predicting the future of elephant flows. To address these issues, we propose a new stochastic performance evaluation model to estimate the loss rate of two state-of-art flow scheduling algorithms including Equalcost multi-path routing (ECMP), Hedera besides a flow congestion control algorithm which is Data Center TCP (DCTCP). Although these algorithms have theoretical and practical benefits, their effectiveness has not been statistically investigated and analyzed in conserving the elephant flows. Therefore, we conducted extensive experiments on the fat-tree data center network to examine the efficiency of the algorithms under different network circumstances based on Monte Carlo risk analysis. The results show that Hedera is still risky to be used to handle the elephant flows due to its unstable throughput achieved under stochastic network congestion. On the other hand, DCTCP found suffering under high load scenarios. These outcomes might apply to all data center applications, in particular, the applications that demand high stability and productivity. Index Terms-Elephant flow, SDN, Risk analysis, Value-at-Risk, Flow scheduling, Congestion control. I. INTRODUCTION Nowadays, many enterprises leverage data center fabrics to manage highly-demanded bandwidth applications. Applications like Hadoop [1] and MapReduce [2] rely on hundreds or thousands of servers to provide high availability and scalability; therefore large data is transferred through the data center network to achieve these requirements. However, other types of data center applications such as regular web services are hosted inside the data center as well, due to the guaranteed availability and reliability. Because of these substantial requirements, many data center topologies evolved like hyperx [3], flattened butterfly [4], and fat-tree [5]. On the other hand, many traffic management techniques emerged, like throughput-based forwarding and load balancing [6]. Typically, the applications of data center produce two types of flows which are mice and elephant flows [6]. Mice flows are known as the smallest and shortest-lived TCP flows in the network and more sensitive to the communication delay. Whereas the most massive and longlived TCP flows, elephant flows, are more affected by the residual link bandwidth [6].

Single-Instruction-Multiple-Data Instruction-Set-Based Heat Ranking Optimization for Massive Network Flow

“…Li et al proposed a new Sketch algorithm, WavingSketch, which can be applied to four applications: finding top-k frequent items, finding top-k heavy changes, finding top-k persistent items, and finding top-k Super-Spreaders [9]. Alawadi A H et al proposed a new stochastic performance evaluation model to detect and mark top-k heat flows, and then optimize these flows using dynamic load balancing, equal-cost multi-path routing (ECMP), and other traffic engineering technologies, which can improve throughput and flow completion time (FCT) [10]. L. Tang et al presented the MV-Sketch algorithm, a fast and compact invertible Sketch algorithm that supports heavy flow detection with small and static memory allocation [11].…”