tive Network technology envisions deployment of virtual execution environments within network elements, such as switches and routers, so that nonhomogeneous processing can be applied to network traffic associated with services, flows, or even individual packets. To use such a technology safely and efficiently, individual nodes must provide mechanisms to enforce resource limits associated with specific network traffic. In order to provide enforcement mechanisms, each node must have a meaningful understanding of the resource requirements for specific network traffic. In Active Network nodes, resource requirements typically come in three categories: bandwidth, memory, and processing. Well-accepted metrics exist for expressing bandwidth (bits per second) and memory (bytes) in units independent of the capabilities of particular nodes. Unfortunately, no well-accepted metric exists for expressing processing (i.e., CPU time) requirements in a platform-independent form. This paper investigates a method to express the CPU time requirements of Active Applications (similar to distributed, mobile agents) in a form that can be meaningfully interpreted among heterogeneous nodes in an Active Network. The model consists of two parts: a node model and an application model. For modeling applications, the paper describes and evaluates a semi-stochastic state-transition model intended to represent the CPU usage requirements of Active Applications. Using measurement data, the general model is instantiated for two Active Applications, ping and multicast. The instantiated models are simulated, and the simulation results are compared against real measurements. For both Active Applications, the simulated and measured CPU time usage compare within 5 % for the mean and for high percentiles. The paper also evaluates three different scaling factors that might be used to transform a model accurate on one node into terms that prove accurate on another node. We found that scaling yields inaccurate results when based on the ratio of processor speeds or on the ratio of performance on a preliminary node calibration workload. When we used an arbitrary scaling factor to achieve a close correspondence between simulation and real measurements for the ping application, that same scaling factor proved effective for transforming a model of the multicast application.
Citation: Proceedings of the 2nd International Workshop on Software Performance (WOSP 2000)
Pub Type: Journals
active networks, benchmarks, resource management, software performance metrics, state-transition models