On the impact of variability on the buffer dynamics in IP networks

Youngmi Joo (jym@leland.stanford.edu)
Electrical Engineering, Stanford University

Vinay Ribeiro (vinay@rice.edu)
Electrical and Computer Engineering, Rice University

Anja Feldmann (anja@research.att.com)
AT&T Labs-Research, Florham Park, NJ

Anna Gilbert (agilbert@research.att.com)
AT&T Labs-Research, Florham Park, NJ

Walter Willinger (walter@research.att.com)
AT&T Labs-Research, Florham Park, NJ

The main objective of this paper is to demonstrate in the context of a simple TCP/IP-based network that depending on the underlying assumptions about the inherent nature of the variability of network traffic, very different conclusions can be derived for a number of well-studied and apparently well-understood problems in the areas of traffic engineering and management. For example, by either fully ignoring or explicitly accounting for the empirically observed variability of network traffic at the source level, we provide detailed ns-2-based simulation results for two commonly-used traffic workload scenarios that can give rise to fundamentally different buffer dynamics in IP routers. We also discuss a set of ns-2 simulation experiments to illustrate that the queuing dynamics within IP routers can be qualitatively very different depending on whether the observed variability of measured network traffic over small time scales is assumed to be in part endogenous in nature (i.e., due to TCP's feedback flow control mechanism, which is "closed loop") or is exogenously determined, resulting in an "open loop" characterization of network traffic arriving at the routers.