Multiple source, multiple destination network tomography
Authors:
- M. RabbatMRM. Rabbat
- R. NowakRNR. Nowak
- M. CoatesMCM. Coates
- 23 December 2004
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 3, 1628-1639
Abstract
The problem of identifying topology and inferring link-level performance parameters such as packet drop rate or delay variance using only end-to-end measurements is commonly referred to as network tomography. This paper describes a collaborative framework for performing network tomography on topologies with multiple sources and multiple destinations, with- out assuming the topology to be known. Using multiple sources potentially provides a more accurate and refined characterization of the internal network. We present a novel multiple source active measurement procedure using a semi-randomized probing scheme and packet arrival order measurements which do not require precise synchronization between the participating hosts. A decision-theoretic framework is developed enabling the joint characterization of topology and internal performance. We design a statistical test based on the Generalized Likelihood Ratio Test and Wilks' Theorem. The test quantifies the tradeoff between network topology complexity and performance estimation, and identifies when measurements made by the two sources can be combined to achieve reduced variance performance estimates. The performance and efficacy of the algorithm are assessed through ns-2 simulations and experiments over the Internet. Method Keywords— Statistics, Network measurements, Simu- lations, Experimentation with real networks/testbeds. I. NETWORK TOMOGRAPHY Assessing and predicting internal network behavior is of fundamental importance in a variety of problems such as routing optimization, network management, and anomaly de- tection. However, acquiring direct internal measurements from all parts of the network is not practical due to the distributed nature of the Internet. Additionally, one cannot rely on internal network elements to respond with special purpose messages (i.e. ICMP timestamp exceeded) due to growing security con- cerns. Those who do have access to internal measurements are nearly always restricted from sharing the data for proprietary and privacy reasons. For the purpose of network management, direct link-level measurements are critical for the low-level analysis of equip- ment conditions. However, traditional fault alarms are only triggered after failures occur, and passive measurements gen- erate large amounts of data which are not easily processed online. Ciavattone et al. describe a practical system using ac- tive end-to-end measurements to augment traditional network operations measurements allowing them to proactively detect impairments and react quickly to performance degradation (1). The problem of inferring internal network characteristics using end-to-end measurements is commonly referred to as network tomography, drawing an analogy to the medical to- mography problem of imaging the internals of the human body in a non-intrusive manner (2, 3). Active measurement tech- niques have been designed using both unicast and multicast measurements to estimate link-level performance parameters such as loss rate and delay variance (4-6), in addition to identifying topology (7-9). This paper presents a study of the multiple source, multi- ple destination network tomography problem. Using multiple sources in the context of network tomography, it is possible to identify segments within a network shared by the paths connecting multiple sources and destinations. This information may be useful for identifying potential bottlenecks. Sharing statistics between sources may also be useful for optimizing the use of network resources when transferring large amounts of data. Additionally, in some cases it is possible to fuse information gleaned from multiple sources to get a more accurate and refined network characterization.Keywords
This publication has 15 references indexed in Scilit:
- Predicting Internet network distance with coordinates-based approaches2003
- Merging logical topologies using end-to-end measurements2003
- Inferring link loss using striped unicast probes2002
- Multicast inference of packet delay variance at interior network links2002
- Internet tomographyIEEE Signal Processing Magazine, 2002
- Multicast topology inference from measured end-to-end lossIEEE Transactions on Information Theory, 2002
- Measuring packet reordering2002
- IDMaps: a global Internet host distance estimation serviceIEEE/ACM Transactions on Networking, 2001
- Advances in network simulationComputer, 2000
- The Large-Sample Distribution of the Likelihood Ratio for Testing Composite HypothesesThe Annals of Mathematical Statistics, 1938