Dynamic Traffic Engineering Considering Service Grade in Integrated Service Network

Abstract

This study proposes a dynamic traffic engineering method to improve network resource utilization efficiency without affecting important services in an integrated service network where multiple grades of service are accommodated. Network resource utilization is biased over time because future fluctuations in traffic demand are unpredictable. Although a service network provider prefers to re-optimize network resource utilization using a dynamic traffic engineering, it is often superfluous for users who do not want their mission-critical communications, for example, executive web meetings or online surgery, to be affected. Therefore, we classified traffic into very important packets and commoners, and only commoners were dynamically optimized. To classify traffic, we assigned the identifier to a packet on the application side. In this approach, only the application and edge nodes of the network required modification; the core network routers were unaffected. Furthermore, we established the architecture of the edge node using a software-defined network controller and P4 switch. We also provided an algorithm for optimization based on linear programming and demonstrated that it is possible to improve network resource utilization efficiency without affecting very important packets and that the computational and memory costs are practical.

Spatial anomaly detection using fast xFlow proxy for nation-wide IP network

“…Once individual flows are observed in fine-grained by the Fast xFlow Proxy, an analysis of whether the current behavior of the flow is similar compared to its normal state can be performed. We demonstrated that this simple autoregressive analysis based on correlation values can detect anomalies with high accuracy [4].…”

“…We proposed the Fast xFlow Proxy, which enables finegrained measurement of flows in large-scale wide-area IP networks [3], and devised an algorithm to detect anomalies such as service failures and rapid increases in traffic caused by cyber-attacks [4]. The results from flow measurement protocols are usually presented as macroflows superimposed of various communication flows.…”

Network Anomaly Detection Through IP Traffic Analysis With Variable Granularity

“…Recently, efficient methods based on probabilistic data structures have been proposed, such as flow radar [12] using a Bloom filter, MV-sketch [13] using a count-min sketch structure, and HashFlow [14], which identifies elephant flows and focuses on their observations. By observing communication flows using these methods, network operators can properly assess service failures or apply traffic engineering [5] to their networks.…”

“…5(d), the anomaly is difficult to detect because it occurs only in a portion of the observed traffic. Using this time-series traffic data as input, the correlation was computed as an evaluation index using (5) in Section IV.B. noise components at the rates of 1%, 5%, and 30%, respectively.…”

“…Although the xFlow protocol can practically collect flow information over the Internet, it has limitations in largescale wide-are IP networks: packet headers are encapsulated by additional headers assigned by communication carriers, and user traffic behavior is observed in a more macroscopic state. For example, encapsulation by the layer 2 tunneling protocol (L2TP) [3] is used for PPPoE user authentication, and multiple labels by segment routing using multiprotocol label switching (SR-MPLS) [4] are added as outer headers for virtual private network (VPN) users or flexible network control by traffic engineering [5]. Consequently, the xFlow protocol does not obtain detailed per-flow information but rather more macroscopic information from the outer headers.…”