Internet Mapping: from Art to Science May 25, 2009

(i) Three most important things

1. We critically depend on the Internet for our professional, personal, and political lives but we know little about what keeps the Internet stable as the Internet becomes continuously challenging to research and analyze. The paper presents the design of an infrastructure and operating system platform known as Ark that supports large-scale active measurements studies of the global Internet for Internet topology mapping.

2.  Easy development and rapid prototyping are important factors in how they promote discovery. A researcher can explore more risky ideas which could have higher returns, by lowering the cost needed in time and effort to implement a measurement idea. Ark supports rapid prototyping by promoting software development at a high-level of abstraction using dynamic scripting languages and pre-built API’s and services.

3. It should be easy for researchers to use and to build upon the work of others at the granularity of services. Ark supports measurement services by providing a tuple space which acts as the unified mechanism for transport and messaging. A user can easily deploy a measurement service by writing a program that interprets tuples as commands, performs some measurement, and returns the result as a tuple.

(ii) Most glaring problem

The most glaring problem would be that paper doesn’t discuss much how Ark has helped study Internet topology. The paper mentions a couple researches that have implemented Ark but doesn’t really provide any actual results and conclusions.

(iii) Future Research Directions

Future research directions for this work would be to have Ark implemented by other research communities so that more data can be collected on Internet topology and expand Ark to perform more IPv6 topology measurements.

Internet Mapping: from Art to Science May 25, 2009

1. Distributed and scalable measurement architecture

The paper suggests a measurement/monitoring architecture that allows different groups of users dispersed over geographically separated locations to make topology measurements. It also provides a common, shared database that is updated with all measurements made. This way, not only is support for a large user base provided, but also, diverse and time skewed measurements of the Internet topology is  obtained. It makes use of tuple-space co-ordination model to achieve this shared memory structure. Decentralized measurement also introduces randomness into measurements which more accurately captures topology data.

2. Incremental development

The architecture makes it feasible for the users to build upon the work and development done by others. It shields the users from underlying complexities and allows them to use a scripting language to develop their codes for the measurement.

3. Interpreting and processing gathered information

The paper suggests some interesting methods for deriving topology information from the raw data gathered. For eg. they mention a technique to obtain the IP addresses of hosts connected to a router from data obtained by measuring paths to a list of IP addresses covering all /24 prefixes in an address space.

Short-comings

A shared memory to update measurements and a common code to which any user can make additions mean a faulty or malicious code and compromise the entire structure. However, this is a trade-off one needs to make with any form of open-ware. In addition, there is no mention of accommodating for dynamic and changing network topologies while allowing different users to update the database with measurements made at different instances of time. The method of alias resolution mentioned by them seems to only support end routers with hosts directly connected to the router of concern.

Future Work

The idea is still young and can grow in multiple dimensions. The API can be developed further to provide for greater ease 0f use and deployment. Increased functionality can be added to their data interpretation tools to glean greater insight into the network topology from the raw data gathered by the system. Attempt can be made to increase the variety of data gathered by the system as well.

Internet Mapping: from Art to Science May 25, 2009

The authors describe a distributed network measurement infrastructure, Archipelago (Ark), which they have deployed around the world to help researchers coordinate large scale Internet measurements. The Ark infrastructure is deployed at over 31 sites around the globe and can coordinate using a tuple-based directory service. This is the primary contribution of this paper.

Ark is deployed as individual monitors (nodes) that run the Ark software. The interface is written in Ruby, though lower level access using C/C++ is supported as well. The distributed directory service support is based on Marinda, a tuple-space distributed shared memory. Monitors can modify this tuple-space to implement support for dynamic coordinate and even measurement services. The tuple-space serves as the layer of abstraction that services can use to read/write domain/range values.

As described, such an infrastructure by itself is not such a new or useful tool. However, the designers wrap the implementation with an easy to use API (in Ruby). This allows measurement studies to be written in a high level scripting language, which is conducive to rapid development and high adoption. Furthermore, the designers decided to allow access to anyone who wants it. No special authorization is needed.

Using this system, the authors have been able to gather Internet topology information that is annotated with link latencies, router ids, and AS numbers (whenever possible). This well annotated global topology can be used by many researchers for macroscopic studies. The authors claim that using Ark, they have been able to coordinate with other researchers and develop improved methods for such measurement tasks as: reconstructing router level topology, dns resolution, and ICMP & UDP topology probing.

One of the most attractive features of Ark is that it is open to anyone who wishes to gather measurement data. The paper describes the Ark infrastructure as secure, but there is not explanation of how it is secure. Indeed, if anyone can use the system, access the tuple-space shared memory to coordinate with other services/researchers, and there is no authorization, how is the system secure? This is not addressed in the paper. Some discussion of the security claim is in order.

I’d expect to see further development for Ark in the direction of built-in measurement services. It seems like the addition of a library of highly optimized (best practice) basic measurement routines would be very useful for any researcher starting out with a measurement project.