Title : A Case for Testbed Embedding Service

Authors : Jeffrey Considine, John W. Byers, Ketan Mayer-Patel (UNC)

Abstract: 

Today's networking community is becoming increasingly skeptical of the
significance of research results founded wholly upon experimental
results conducted in simulation. Now, with the availability of
wide-area distributed testbeds such as PlanetLab, it is feasible to
move beyond evaluation by simulation, and to perform wide-area
experiments across the Internet as an alternative. However, one must
reconcile a major challenge with this approach: while use of a
distributed testbed affords much greater realism than a network
simulation, there is a significant downside, as tight control over
one's experiments is relinquished.

We argue that providing services for distributed testbeds that capture
aspects of the specifiable, repeatable behavior implicit in simulation
and emulation will be an integral component of next-generation
testbeds.  Designing these services poses fundamental and technically
difficult Computer Science challenges.  As a case study, we consider
the following embedding problem: given a desired end-system topology
consisting of a set of pairwise constraints (such as upper and lower
bounds on bandwidth and delay), locate a representative subtopology
within a wide-area testbed that satisfies those constraints.  We
outline the theoretical foundations of a distributed service
encompassing adaptive network measurements and distributed search that
produces such an embedding, and report on preliminary experimental
results conducted on PlanetLab.