Kuai Xu

A first look at inter-data center traffic characteristics via Yahoo! datasets

Effectively managing multiple data centers and their traffic dynamics pose many challenges to their operators, as little is known about the characteristics of inter-data center (D2D) traffic. In this paper we present a first study of D2D traffic characteristics using the anonymized NetFlow datasets collected at the border routers of five major Yahoo! data centers. Our contributions are mainly two-fold: i) we develop novel heuristics to infer the Yahoo! IP addresses and localize their locations from the anonymized NetFlow datasets, and ii) we study and analyze both D2D and client traffic characteristics and the correlations between these two types of traffic. Our study reveals that Yahoo! uses a hierarchical way of deploying data centers, with several satellite data centers distributed in other countries and backbone data centers distributed in US locations. For Yahoo! US data centers, we separate the client-triggered D2D traffic and background D2D traffic from the aggregate D2D traffic using port based correlation, and study their respective characteristics. Our findings shed light on the interplay of multiple data centers and their traffic dynamics within a large content provider, and provide insights to data center designers and operators as well as researchers.

Improving VoIP quality through path switching

The current best-effort Internet cannot readily provide the service guarantees that VoIP applications often require. Path switching can potentially address this problem without requiring new network mechanisms, simply by leveraging the robustness to performance variations available from connectivity options such as multi-homing and overlays. In this paper, we evaluate the effectiveness and benefits of path switching in improving the quality of VoIP applications, and demonstrate its feasibility through the design and implementation of a prototype gateway. We argue for an application-driven path switching system that accounts for both network path characteristics and application-specific factors (e.g., codec algorithms, playout buffering schemes). We also develop an application path quality estimator based on the ITU-T E-model for voice quality assessment, and an application-driven path switching algorithm that dynamically adapts the time scales over which path switching decisions are made to maximize voice quality. Through network emulation and experiments over a wide-area multi-homed test bed, we show that, with sufficient path diversity, path switching can yield meaningful improvements in voice quality. Hence by exploiting the inherent path diversity of the Internet, application-driven path switching is a viable option in providing quality-of-service to applications.

Internet traffic behavior profiling for network security monitoring

Recent spates of cyber-attacks and frequent emergence of applications affecting Internet traffic dynamics have made it imperative to develop effective techniques that can extract, and make sense of, significant communication patterns from Internet traffic data for use in network operations and security management. In this paper, we present a general methodology for building comprehensive behavior profiles of Internet backbone traffic in terms of communication patterns of end-hosts and services. Relying on data mining and entropy-based techniques, the methodology consists of significant cluster extraction, automatic behavior classification and structural modeling for in-depth interpretive analyses. We validate the methodology using data sets from the core of the Internet.

Exploring the performance benefits of end-to-end path switching

This work explores the feasibility of improving the performance of end-to-end data transfers between different sites through path switching. Our study is focused on both the logic that controls path switching decisions and the configurations required to achieve sufficient path diversity. Specifically, we investigate two common approaches offering path diversity multi-homing and overlay networks - and investigate their characteristics in the context of a representative wide-area testbed. We explore the end-to-end delay and loss characteristics of different paths and find that substantial improvements can potentially be achieved by path switching, especially in lowering end-to-end losses. Based on this assessment, we develop a simple path-switching mechanism capable of realizing those performance improvements. Our experimental study demonstrates that substantial performance improvements are indeed achievable using this approach.

Diffusive Logistic Model Towards Predicting Information Diffusion in Online Social Networks

Online social networks have recently become an effective and innovative channel for spreading information and influence among hundreds of millions of end users. Most of prior work either carried out empirical studies or focus on the information diffusion modeling in temporal dimension, little attempt has been given on understanding information diffusion over both temporal and spatial dimensions. In this paper, we propose a Partial Differential Equation (PDE), specifically, a Diffusive Logistic (DL) equation to model the temporal and spatial characteristics of information diffusion. We present the temporal and spatial patterns in a real dataset collected from a social news aggregation site, Digg, and validate the proposed DL equation in terms of predicting the information diffusion process. Our experiment results show that the DL model is able to characterize and predict the process of information propagation in online social networks. For example, for the most popular news with 24,099 votes in Digg, the average prediction accuracy of DL model over all distances during the first 6 hours is 92.08%. To the best of our knowledge, this paper is the first attempt to use PDE-based model to study the information diffusion process in both temporal and spatial dimensions in online social networks.

On Properties of Internet Exchange Points and Their Impact on AS Topology and Relationship

Internet eXchange Points (IXPs) are one of two primary methods for Autonomous Systems (ASes) to interconnect with each other for exchanging traffic and for global Internet reachability. This paper explores the properties of IXPs and their impact on the AS topology and AS business relations using Scriptroute and Skitter traceroute probes, BGP routing archives and other data. With these datasets we develop an algorithm to discover IXPs and infer ASes that participate at these IXPs. Using the discovered IXPs and their inferred AS participants, we analyze and characterize the properties of IXPs and their participants such as size, geographical locations. We also investigate the impact of IXPs on the global AS topology and business relations between ASes. Our study sheds light on the Internet interconnection practices and the evolution of the Internet, in particular, the potential role IXPs play in such evolution.

On positive influence dominating sets in social networks

In this paper, we investigate the positive influence dominating set (PIDS) which has applications in social networks. We prove that PIDS is APX-hard and propose a greedy algorithm with an approximation ratio of H(@d) where H is the harmonic function and @d is the maximum vertex degree of the graph representing a social network.

Characterizing Information Diffusion in Online Social Networks with Linear Diffusive Model

Mathematical modeling is an important approach to study information diffusion in online social networks. Prior studies have focused on the modeling of the temporal aspect of information diffusion. A recent effort introduced the spatiotemporal diffusion problem and addressed the problem with a theoretical framework built on the similarity between information propagation in online social networks and biological invasion in ecology [1]. This paper examines the spatio-temporal characteristics in further depth and reveals that there exist regularities in information diffusion in temporal and spatial dimensions. Furthermore, we propose a simpler linear partial differential equation that takes account of the influence of spatial population density and temporal decay of user interests in the information. We validate the proposed linear model with Digg news stories which received more than 3000 votes during June 2009, and show that the model can describe nearly 60% of the news stories with over 80% accuracy. We also use the most popular news story as a case study and find that the linear diffusive model can achieve an accuracy as high as 97:41% for this news story. Finally, we discuss the potential applications of this model towards finding super spreaders and classifying news story into groups.

Positive Influence Dominating Set in Online Social Networks

Online social network has developed significantly in recent years as a medium of communicating, sharing and disseminating information and spreading influence. Most of current research has been on understanding the property of online social network and utilizing it to spread information and ideas. In this paper, we explored the problem of how to utilize online social networks to help alleviate social problems in the physical world, for example, the drinking, smoking, and drug related problems. We proposed a Positive Influence Dominating Set (PIDS) selection algorithm and analyzed its effect on a real online social network data set through simulations. By comparing the size and the average positive degree of PIDS with those of a 1-dominating set, we found that by strategically choosing 26% more people into the PIDS to participate in the intervention program, the average positive degree increases by approximately 3.3 times. In terms of the application, this result implies that by moderately increasing the participation related cost, the probability of positive influencing the whole community through the intervention program is significantly higher. We also discovered that a power law graph has empirically larger dominating sets (both the PIDS and 1-dominating set) than a random graph does.

A first step toward understanding inter-domain routing dynamics

BGP updates are triggered by a variety of events such as link failures, resets, routers crashing, configuration changes, and so on. Making sense of these updates and identifying the underlying events is key to debugging and troubleshooting BGP routing problems. In this paper, as a first step toward the much harder problem of root cause analysis of BGP updates, we discuss if, and how, updates triggered by distinct underlying events can be separated. Specifically, we explore using PCA (Principal Components Analysis), a well known statistical multi-variate technique, to achieve this goal.We propose a method based on PCA to obtain a set of clusters from a BGP update stream; each of these is a set of entities (either prefixes or ASes) which are affected by the same underlying event. Then we demonstrate our approach using BGP data obtained by simulations and show that the method is quite effective. In addition, we perform a high level analysis of BGP data containing well known, large scale events.