Calibrating an Active Network Node

Active Network technology envisions deployment of virtual execution environments within network elements, so that nonhomogeneous processing can be applied to network traffic. For management purposes, each node must have a meaningful understanding of resource requirements in terms of bandwidth, memory, and processing. To express the processing requirements in a platformindependent manner, we are developing a model of CPU time usage, which comes in two parts: a node model and an application model. In order to generate instances of the model, one needs to gather some metrics relative to the platform, that is, to calibrate a node. We have investigated what factors this process of calibration should account for, and especially how background load on a node affects our ability to obtain accurate calibrations for the CPU time used by node operating system calls and by virtual execution environments. We have shown that a background load, either computation intensive or input/output intensive, has little influence on the calibration. On the contrary, a memory consuming background load introduces an overhead in some measurements. The paper draws the conclusion that the calibration of a node can be done whatever the background load, provided that the memory consuming loads can be suppressed if necessary.