| 11 | | |
| 12 | | In order to perform realistic network simulations, one needs a traffic generator that is capable of generating realistic synthetic traffic in a closed-loop fashion that "looks like" traffic found on an actual network. |
| 13 | | |
| 14 | | The Tmix system takes as input a packet header trace file captured from a network link of interest (such as the link between the UNC campus and the rest of the internet). This trace is reverse-compiled into a connection vector (or cvec) file, which is a source-level characterization of each TCP connection present in the trace. Tmix then uses this information to emulate the socket-level behavior of the source application that created the corresponding connection in the trace. The resulting traffic generation is statistically representative of the traffic measured on the real link. |
| 15 | | |
| 16 | | == Traffic Generation == |
| 17 | | One of the most complex components of empirical evaluations is modeling and generating |
| 18 | | realistic Internet traffic. The mix of the ever changing and varied applications that constitute the |
| 19 | | actual Internet traffic makes this a daunting task. Moreover, Internet traffic is different when |
| 20 | | sampled at different times and in different parts of the globe. Networking researchers have |
| 21 | | grappled with this problem by taking snapshots of Internet traffic at different times and at various |
| 22 | | points in the network, and modeling the same for generating traffic in the lab. The generally held 3 |
| 23 | | belief is that the more realistic the traffic used, the more reliable are the results of the empirical |
| 24 | | evaluations using that traffic. Practice, however, does not adhere to this principle. So, although |
| 25 | | laboratory testbeds and methods for simulations have evolved over the years, the question about |
| 26 | | what constitutes essential components for modeling realistic traffic remains open for debate. For |
| 27 | | example, networking researchers agree that realistic traffic generation for empirical research is |
| 28 | | best accomplished by capturing traffic on a production link and then using source-level models to |
| 29 | | generate this traffic in the laboratory or simulator. Source-level models capture the application |
| 30 | | exchanges and application behavior on the ends (sources) of the TCP connections. But how do |
| 31 | | you go from the original captured traffic to an acceptable source-level model? Which of the |
| 32 | | several measures derived from the traffic sources should you model in your workload for your |
| 33 | | experiments? Would your modeling choices for traffic generation impact the outcome of your |
| 34 | | experiments? If yes, how significant would the impact be? These remain open questions. |
| | 11 | The Tmix system takes as input a TCP/IP packet header trace file captured from a network link of interest. This trace is |
| | 12 | reverse-compiled into a connection vector (or cvec) file, which is a source-level characterization of each TCP connection present in the trace. Tmix then uses this information to emulate the socket-level behavior of the source application that created the corresponding connection in the trace. The resulting traffic generation is statistically representative of the traffic measured on the real link. |
