The paper "A Measurement-Based Analysis of Multihoming" proposed by Akella et al. at SIGCOMM'03 shows different aspects of performances improvements in multihomed environments.

The journal article we discussed in [08-03-17] is basically the same as this paper, so that we only present a fast summary of the paper.

First, the authors presents the problem of performances improvements due to multihoming with a small data set and show that there is something interesting that needs more investigation.

So that, they propose to estimate the turnaround time improvement from an enterprise point of view by passing from a 1-homed site to a k-multihomed site. They show that having a second ISP is interesting on average for the turnaround time. But even if the addition of other ISPs improves the turnaround time on average, the improvement is marginal for 4-multihoming or more.

From a web-server point of view, they also show that increasing the number of ISPs improves the "web transfer response time" but once again, adding a new ISP become marginal after 4 providers (even before). In addition, in Figure 12, we can observe that the major proportion of traffic will be transferred through the 2 or 3 best ISPs even in 9-multihoming.

The paper also presents a discussion about improvement of path diversity (not the reliability as traceroutes only gives IP information) in k-multihoming. Multihoming can improve path diversity by 25% or more. To estimate the improvement of path diversity, the authors propose two metrics R_1 and R_2. (section 5.1) It is important that the dataset is limited to one direction diversity.

The discussion at the end of the paper shows that accurate and timely estimators are The Key to obtain performances improvements. Worse, "stale information could lead to selecting sub-obptimal routes".

This paper can be considered as a really good reference for our IDIPS project: http://inl.info.ucl.ac.be/idips

The paper is available at http://pages.cs.wisc.edu/~akella/papers/multihoming.pdf