在NCTUns網路模擬器支援平行模擬

Abstract

對發展以及診斷網路協定的研究人員而言，以軟體實做的網路模擬器是相當有價值的工具。對某些中小型的網路來說，模擬足已能夠洞悉這些網路內在的關鍵行為。然而，對於要模擬上千個節點，且每個節點上面都執行數個應用程式的模擬案例，單一機器因為在中央處理器速度以及記憶體空間的限制下，難以完成如此大規模模擬。因此，設計實做平行模擬方法來延伸模擬器的模擬規模是有其價值的。NCTUns為分散式事件驅動的網路模擬器，在本論文中，我們將數種保守法應用平行分散事件模擬於NCTUns網路模擬引擎上，並且比較不同版本保守法在NCTUns網路模擬引擎上執行的效能差別。

本篇論文中，我們首先會描述如何將保守法應用在NCTUns網路模擬器上，然後檢驗這些保守法的效能，接下來比較兩個網路模擬器，NCTUns以及NS2上平行模擬的效能，最後討論一些影響平行模擬效能的因子。

Network simulators implemented in software are valuable tools for researchers to develop and diagnose network protocols. In certain cases, simulations of small to medium sized networks may be sufficient to gain critical insights into the behaviors of those networks. However, for simulation cases with thousands of nodes, each node may have several application programs that need to be run on it. A single machine cannot accommodate the required CPU and memory resources for running such large-scale simulations. Therefore, it is valuable for a simulator to design and implement a parallel simulation methodology to expand its scalability. NCTUns is a network simulator that uses discrete event-driven system. In this thesis, we implemented a parallel discrete event simulation engine using several conservative algorithms for NCTUns and compared the performances of several versions of NCTUns simulation engine.

In this paper, we first describe how conservative algorithms for parallel simulations are applied to the NCTUns network simulator. Then we examine the performances of a set of parallel conservative algorithms. Next, we compare the performances of parallel simulations between two popular network simulators, NCTUns and NS2. Finally, we discuss the effects of several important factors in parallel simulation.