Estimating Traffic Anomalies for Throughput Prediction on Network Virtualization

Network test beds based on virtualization technology have been used in network research. Network measurement is affected by shared and virtualized resources on a node. Oversized packet spacings, which can occur due to CPU scheduling latency, are a major cause of both throughput instability and imprecise network measurement on a virtualized network test bed. These spacings are anomalies in virtualized network environment. Although CPU availability is an important criterion for estimating these anomalies, a naive method for measuring it may over consume CPU resources and affect the performance of other tasks. We observed the resource states of a system during throughput measurement and applied principal component analysis to an approximately 7 × 8000 matrix of the resource states to obtain criteria for estimating the anomalies instead of using the naive method. We show that the top two principal components account for 84% of the original dataset and can describe the original resource state. Component loadings and a scatter plot of the first and second component scores can provide a simple but descriptive enough view of the resource state for anomaly estimation. The first component is workloads on the node, and the second one is lack of resources, leading to anomalies. We determined the appropriate component boundary scores for an anomaly area by using Bayes' approach and used the boundaries to evaluate the anomalies with an input dataset. The results showed that our approach can be used to estimate anomalies.

[1]  Peter A. Dinda,et al.  Characterizing and Predicting TCP Throughput on the Wide Area Network , 2005, 25th IEEE International Conference on Distributed Computing Systems (ICDCS'05).

[2]  Richard Wolski,et al.  The network weather service: a distributed resource performance forecasting service for metacomputing , 1999, Future Gener. Comput. Syst..

[3]  Richard Wolski,et al.  Multivariate Resource Performance Forecasting in the Network Weather Service , 2002, ACM/IEEE SC 2002 Conference (SC'02).

[4]  KyoungSoo Park,et al.  CoMon: a mostly-scalable monitoring system for PlanetLab , 2006, OPSR.

[5]  Zhi-Li Zhang,et al.  Co-designing the failure analysis and monitoring of large-scale systems , 2008, PERV.

[6]  Paul Barford,et al.  An active measurement system for shared environments , 2007, IMC '07.

[7]  Ian T. Jolliffe,et al.  Discarding Variables in a Principal Component Analysis. I: Artificial Data , 1972 .

[8]  R. Cattell FACTOR ANALYIS: AN INTRODUCTION TO ESSENTIALS. II. THE ROLE OF FACTOR ANALYSIS IN RESEARCH. , 1965, Biometrics.

[9]  Toshio Hirotsu,et al.  Analysis of Anomalies on a Virtualized Network Testbed , 2010, 2010 10th IEEE International Conference on Computer and Information Technology.

[10]  David E. Culler,et al.  Operating Systems Support for Planetary-Scale Network Services , 2004, NSDI.

[11]  Larry L. Peterson,et al.  Experiences building PlanetLab , 2006, OSDI '06.

[12]  Larry L. Peterson,et al.  Using PlanetLab for network research: myths, realities, and best practices , 2005, OPSR.