Flexible Performance Prediction of Data Center Networks using Automatically Generated Simulation Models

Abstract

Using different modeling and simulation approaches for predicting network performance requires extensive experience and involves a number of time consuming manual steps regarding each of the modeling formalisms. In this paper, we propose a generic approach to modeling the performance of data center networks. The approach offers multiple performance models but requires to use only a single modeling language. We propose a two-step modeling methodology, in which a high-level descriptive model of the network is built in the first step, and in the second step model-to-model transformations are used to automatically transform the descriptive model to different network simulation models. We automatically generate three performance models defined at different levels of abstraction to analyze network throughput. By offering multiple simulation models in parallel, we provide flexibility in trading-off between the modeling accuracy and the simulation overhead. We analyze the simulation models by comparing the prediction accuracy with respect to the simulation duration. We observe, that in the investigated scenarios the solution duration of coarser simulation models is up to 300 times shorter, whereas the average prediction accuracy decreases only by 4 percent.

ML-based Performance Prediction of SDN using Simulated Data from Real and Synthetic Networks

“…Consequently, similar benchmarking analyses regarding performance metrics may be conducted if the deployed network can be accessed for performance studies. As this is rarely possible, analogous to controller evaluations, some works leverage simulations [22], while others utilize mathematical models [23], [24].…”

A Taxonomy of Techniques for SLO Failure Prediction in Software Systems

“…Analytic solutions are usually much faster to compute; however, they might not be able to provide the same detail as a simulation or are only applicable to certain scenarios or system types. Therefore, a variety of approaches have been developed to enable dynamically switching between different methods (see, e.g., Walter et al [116,117] for computer systems or Rygielski et al [118,119] for networks) based on the current demand, or alternatively altering the prediction model itself [120,121]. These approaches accomplish this by working on a higher abstraction layer, e.g., the white-box descriptions of Section 5.2.3.…”