Presented at IFIP/Networking

Year

2013

Abstract

One of the challenges that distributed systems designers face is that their performance is very sensitive to the characteristics of the underlying network. Hence, simple models

that accurately describe some statistical properties of the Internet

can be very helpful in the modelling, simulation and design of

Internet-scale distributed systems.

In this paper we present a model for the analysis of Internet

round trip time (RTT) and its relationship to geolocation distance.

This model is based on a novel dataset comprising ∼200 million

RTT samples between ∼54 thousand DNS servers, and can be

used to accurately predict the median RTT between two Internet

hosts separated by a given geographical distance.

Our main contribution is a procedure for the geographic

analysis of RTT that reveals large-scale routing information.

We accomplish this by investigating RTT on the basis of large,

disjoint geographic components. By applying a novel median-

based, least-squares fitting algorithm to traffic flows between

these components, we analyse their RTT\distance behaviour and

compute a large-scale routing excess that quantifies the extra

distance beyond the great circle that packets traverse when they

flow between large-scale geographic areas.

that accurately describe some statistical properties of the Internet

can be very helpful in the modelling, simulation and design of

Internet-scale distributed systems.

In this paper we present a model for the analysis of Internet

round trip time (RTT) and its relationship to geolocation distance.

This model is based on a novel dataset comprising ∼200 million

RTT samples between ∼54 thousand DNS servers, and can be

used to accurately predict the median RTT between two Internet

hosts separated by a given geographical distance.

Our main contribution is a procedure for the geographic

analysis of RTT that reveals large-scale routing information.

We accomplish this by investigating RTT on the basis of large,

disjoint geographic components. By applying a novel median-

based, least-squares fitting algorithm to traffic flows between

these components, we analyse their RTT\distance behaviour and

compute a large-scale routing excess that quantifies the extra

distance beyond the great circle that packets traverse when they

flow between large-scale geographic areas.

Description

This paper describes the collection and analysis of a very large collection of delay data. The data is collected using the Turbo King method (recursive queries of Domain Name Servers). A model is given to fit the data. The data is analysed by continent.

It is shown that there is a strong relationship between delay and Haversine distance (the distance across the surface of the earth). However, this relationship breaks down in specific cases and occasionally, for some continent pairs, the opposite relationship is shown (longer distance is shorter delay).

Preprint

bibtex

@INPROCEEDINGS{rtt_geog_2014,

author={Landa, R. and Araujo, J.T. and Clegg, R.G. and Mykoniati, E. and Griffin, D. and Rio, M.},

booktitle={IFIP Networking Conference, 2013},

title={The large-scale geography of Internet round trip times},

year={2013},

pages={1-9}

}

author={Landa, R. and Araujo, J.T. and Clegg, R.G. and Mykoniati, E. and Griffin, D. and Rio, M.},

booktitle={IFIP Networking Conference, 2013},

title={The large-scale geography of Internet round trip times},

year={2013},

pages={1-9}

}

Paper type

Subject area