Understanding Network Dynamics: The Race is On

While the Internet plays an important role in our day-to-day lives, the global routing infrastructure is surprisingly fragile and the network dynamics of the Internet are not well understood. The RUBINET (Robust & Ubiquitous Networking) research group strives to leverage Internet measurement to characterize the interactions of various network components and across multiple layers. This talk will first give a high-level summary of our finding in terms of Internet routing dynamics. Then we will revisit the design decision of overlay networks to allow routing control at the application layer.

Given the increasing popularity of overlay networks, it is critical to ask the question: can overlay networks and underlying IP networks form a synergistic co-existence? Overlay networks attempt to take control over routing to achieve better performance in the presence of network failures or congestions. ISPs do the same by employing common traffic engineering techniques such as link weight settings, load balancing and routing policies. In this talk, I will illustrate how an uncoordinated effort to the two layers to recover from failures may cause performance degradation and unpredictable behavior from an ISP's view. We also show how current traffic engineering techniques are inadequate to deal with emerging overlay network services.

Multiple similar or dissimilar overlay networks making independent routing decisions could also experience race conditions, resulting in oscillations in both route selection and network load. I will present our analytical model for predicting the synchronization probability of two overlays and shows that it is non-negligible across a wide range of parameter settings, thus implying that the ill-effects of synchronization should not be ignored. The model is validated through simulations that are designed to capture the transient routing behavior of both the IP- and overlay-layers. The effects of factors such as path diversity (measured in round trip times) and probing aggressiveness on these race conditions will be discussed. Finally, I will discuss the implications of our study on the design of overlay networks and the choice of their path probling parameters.