[mat-wg] Thoughts on the Internet Measurement Ecosystem

["Richard L. Barnes" rbarnes@localhost]

Dear MAT WG list,

I'm writing to share with you some thoughts that Ian, Christian, and I have been discussing, about how to expand the RIPE TTM network, better integrate it with other measurement systems around the world, and generally knit the Internet measurement ecosystem into a more cohesive whole.  We're planning to have some discussion on these topics at the upcoming RIPE meeting in Rome, but we wanted to throw this out to the list to give people an introduction and hopefully spark some early discussions.

The network measurement landscape right now is a series of stove-pipes.  RIPE has a measurement network; CAIDA, PlanetLab, and their academic peers have a few different networks; and commercial entities like Akamai, Google, and Skype all collect measurements of the network infrastructure as part of measuring their own products.  Each of these entities collects its own measurements, does its own analysis, and publishes its own results, independent of other measurements going on in the network.

The result of this situation is that Internet measurement systems don't measure the Internet all that well.  The Internet is a huge, diverse place, and each stove-pipe covers only a slice of it. Measurements on PlanetLab cover academic networks pretty well, but not home broadband networks.  Web measurements from Akamai tells us a lot about the web, but little about VoIP or other real-time applications.  Analytics aren't comparable because they're compared on the basis of completely separate data.

So the question arises: How can we as a community work together to create a global network measurement ecosystem that facilitates near-real-time network analytics based on observations from points distributed over a large, diverse portion of the infrastructure?

One could imagine a few possible realizations of this concept, for example:

1. A system for federating measurement systems, allowing them to share measurements in large volumes and in near-real-time.

2. Individual measurement systems and consumers of measurements could join together in something like the DNS OARC, under a common sharing agreement.

3. A central clearing-house for measurements, where some entities throw measurements in, and other entities get notifications when measurements of interest arrive.

(I think of these three examples as analogies to private interconnects, peering points, and route servers, respectively.)  There are bound to be several other paths with different trade-offs, and we welcome suggestions.

There are clearly several interesting problems to solve in the development of this framework.  In addition to the raw technical questions of how much data can be collected and distributed, there are thorny issues around how to incentivize participation, and how to protect the privacy of end users and the anonymity of contributors.  We have some hope that these issues are surmountable, and looks forward to discussions on the list and in Rome.

One final note: This discussion is closely related to the work that the RIPE NCC has been doing to expand their measurement network, e.g., through the RIPE ATLAS project:
<http://labs.ripe.net/Members/dfk/active-measurements-need-more-vantage-points>
What we're going after here is more general, looking at how projects like RIPE ATLAS and the traditional RIPE TTM could integrate more globally.

Thanks for taking the time to read this, and please provide feedback.  Even a simple "Yes, this is great", or "No, go away" would help us gauge whether there's any interest in the community.

Sincerely,
MAT WG chairs
Christian Kaufmann
Ian Meikle
Richard Barnes