Laboratory 9 -

Objectives

1. Use OpNet as a tool to analyze network performance and choose from alternatives in designing networks
2. Investigate the performance of Ethernet hubs vs. switches in more detail than lab 7.

Preparation

My mission this week was to come up with a lab that was a little more interesting than some of the canned "OpNet-by-Numbers" ones. So I tried to come up with a lab where you could put what you've learned about the OpNet simulator to use to solve a network problem.

This time, you'll get a chance to see if I knew what I was talking about with one of the questions on the mid-term, and to see if reality (or a simulation thereof) matches theory. The question was the following:

a. Suppose you want to network a cluster of database servers with several client workstations using standard 10BaseT Ethernet, with 10BaseT hubs. If each of the workstations sends an average of one 1000 byte request to one of the servers each second, and the server responds to each request with a reply that averages 2750 bytes, estimate an upper limit on the number of workstations you could attach to the network before it starts to be overloaded by the traffic. Assume that the database servers can keep up with the load, and you have as many ports on the hubs as you want. (Hint: remember the difference between the theoretical throughput of a 10BaseT network and the practical throughput).
b. Now suppose you get some extra money and can buy Ethernet switches instead of hubs. Would this change your estimated maximum number of workstations? Explain why or why not. If the number would change, would it go up or down?

For the lab, you will use OpNet to confirm or disprove my conclusions on the midterm that you could put 100 nodes on the network and still have it keep up with the load, and that after you changed to switches, you could add more nodes to the network.

In Lab Activities

Boot your machine in Windows and start OpNet. You're starting with a clean slate; feel free to name the project anything you want and set up the network topology the way you like. If you have questions about how to set up a 10BaseT network using hubs, refer back to the Switched LAN experiment.

For the first scenario, verify that the network will really only shovel through about 3 Mbps, regardless how much you try to send. Set up an Ethernet hub - use the ethernet and ethernet_advanced palettes, like you did for lab 7. For the first scenario, set up a hub (I recommend using the ethernet128_hub_adv), and start with 100 workstations. You will need to configure all of the workstations so they generate traffic; for this scenario, rather than trying to add a server, just pretend that each station is going to generate 3750 bytes/sec. A word of warning; if you try to set each station up with packet generation arguments of exponential(1.0) and a packet size of constant(3750), it won't work right - the packet size must be less than 1500, since this is Ethernet. So instead of sending one packet of 3750 bytes each second, you could send three packets of 1250 bytes each per second and get the same average traffic offered to the network.

Select statistics that will let you see how much data is going into the network and how much is coming out, and see what data rate the network can sustain. Run the simulation long enough to make sure that the network has reached a steady state. Create a second scenario with the same equipment, but with 125 workstations, and another scenario with 100 workstations. Does it look like the network is maxed out with 100 workstations, or do 125 work OK? Try increasing the network load even more; instead of increasing the number of workstations, just increase the amount of data each one is trying to send per second. See how the network handles the increased load.

Note that you can speed up the simulation significantly when using a hub by going to the Global Parameters tab in the simulation run window and enabling Eth Hub Optimization. Allegedly it will not affect the collisions or other simulation results, but will just not send a packet out of any ports on the hub other than the one to which it is bound. This reduces the simulation events and the overall time without changing the results.

When I did this experiment, I found that the network was able to send considerably more than 3 Mbps. This is definitely not the expected result. One possibility is that the OpNet model for a hub is acting more like a switch. Create another scenario using a coax Ethernet instead of a 10BaseT Ethernet (like you did for lab 5). Determine what sort of network throughput this network can sustain, using 100 nodes like you tried with the hub. Is it able to send more than 3 Mbps on average?

Now try a scenario with a switch instead of a hub (try the ethernet128_switch) and 100 nodes. Try three different values for the interarrival time and see how much data the network can transmit through the switch. Explain the difference between this and the hub. Note that this is not exactly like the question asked on the test, because the data is not being funneled to a small set of servers, but it's a reasonable first step.

Lab Report

Your lab report should include the following information:

• A description of the networks you simulated and the different scenarios you ran
• A brief description of the statistics you captured for each scenario and the parameters you used when running the simulation (for example, what was the duration of the simulation)
• The graphs you captured from the simulations
• Any conclusions you draw from the experiment

Make sure that you are following Anthony's submission guidelines. Write up a reasonable lab report. As you discovered last week, you'll need to save the report as a PDF file so it is a reasonable size.

The lab write-up is due by the end of the day Wednesday (i.e. 11:59 PM) for the Tuesday lab section and by the end of the day Friday for the Thursday lab section.