ACQUA: Application for Collaborative Estimation of QUality of Internet Access

· Theoretical notions about the tool

· Download

Monitoring the quality of Internet access by active probing

This project is enforced by INRIA and the French National project ANR CMON on Collaborative Monitoring, and consists on a tool that lets the user have an estimation of the anomalies of the Internet based on active measurements of RTT metrics among a predefined set of landmarks (i.e. test points). When an anomaly is detected it is expressed in terms of how many destinations are affected by this anomaly, and how important in terms of RTT variation is this anomaly for these affected destinations.

chadi-w100

Background

Internet is a set of distributed Autonomous Systems that administrate different sections of the overall network. Each of them knows exactly what is happening in its own network, but they do not always provide this information to end-users. This lack of transparency happens due to the fact that in a competitive environment among ISPs, one failure in one of them represents users moving to other provider. So end-users that look for details on performance degradation on the overall network cannot trust in ISP statistics. What choice do they have then? They can collaborate together to get information on each ISP. By doing so, they have a general objective idea on the behavior of the network with regards to each user's point of view.

For this reason projects like Grenouille have arrisen, to provide end-users some more details about the performance on ISPs. Grenouille provides three main metrics about each ISP: upload capacity, donwload capacity and delay measured inside an ISP network between to hosts that are physically far enough to avoid local networks.

The novelty of this project is to introduce a new metric that describes performance degradations on the Internet. This metric is known as Impact Factor, and represents the impact in term of affected destinations that will be perceived by one end-user when facing an anomaly on the Internet’s network. We focus on the paper [1], which provides a practical way of estimating the Impact Factor of failures in a network by doing active measurements. The specific objective of this project is to provide to the user with a tool to obtain this information.

Theoretical notions about the tool

The Impact Factor [1] from the point of view of one end-user represents the seriousness of an anomaly in the network, defined in terms of the amount of destinations that will manifest a service degradation due to this anomaly. An anomaly can be expressed as a variation of the RTT value towards certain destination, i.e. there is an anomaly in the network if one measured RTT value falls outside one expected range, defined by previous samples of RTT.

In the Figure a failure in link A-D might lead to an Impact Factor of 1/3 (case when the vantage point V only uses link A-D to reach F, and other landmarks are not reached through this link). It could be the case that having the same failure (link A-D) the Impact Factor equals 2/3 (case when A-D is used to reach both F and G). Note that it might be the case that a failure in link A-D has impact 0, when this link is not being used by any of the current routes for V. The impact factor ranges between 0 and 1, where 0 means that no destination is affected, and 1 means that there are anomalies in the network such that the possibility to reach all destinations suffers performance degradation.

The Impact Factor defines how many landmarks are affected by the anormaly. To give details about how important these anomalies are, we provide the Shift of the RTT for all the landmarks whose reachability suffered the service degratation. Here an example to understand this. Having a 30% value for the Impact Factor we know that 30% of our landmarks will suffer an anormaly. What do we expect then when trying to reach them? If the Shift of the RTT (for this 30% of landmarks) is 100ms, it tells us that with 30% of probability we will try to reach a landmark whose path suffers an anomaly, and in case we are inside that 30%, the expected value of the RTT for this landmark will differ in 100ms in comparison with the historical value.

The paper[1] shows the minimum amount of probing destinations to analyze in order to have a good estimator of the impact factor as a function of the confidence interval and the significance level that are required for the estimator. Theoretically when the impact factor is completely unpredictable the amount of landmarks required (probe destinations) is large. But in fact it is mentioned that in practice usually two things may happen: some close link may fail resulting in a close to 1 impact factor value (suppose V-A is used to reach the three destinations in Figure 2). Otherwise a far/medium distance link may fail, and this could unlikely affect near half of the destinations. So, in most of the cases for the real topology of Internet we will be facing impact factors close either to 0 or to 1, not in the middle. This implies that the amount of landmarks required to have a good estimator is reduced to just a few, which makes this approach feasible in practice.

This project provides a tool that obtains an estimator of the Impact Factor by taking advantage of the fact that few paths to the destinations are more significant than others, so that they give information enough to have a good approximation on the metric.