I was curious about statistics and network performance at DreamHack and wanted to get statistics for all of the user ports on the local network. There was no way a standard statefull program would be able to get the statistics for the hundreds of switches and thousands of ports on the network.
So I created two simple python scripts. One that would spew out SNMP packets as fast as possible and one that would listen on incoming replies and print out the content on STDOUT.
The sending script created a raw network socket to be able to spoof the source port of the SNMP packet, the source port was also the same as the switch-port (with an offset).
The receiving script used the source IP to know which switch the data came from, and use the network-port to identify which switch-port the data was for. You could scale the setup by having multiple receiving scripts on multiple hosts all saving the incoming data. Remember, SSDs was quite uncommon still.
The result
It worked as intended, almost.
The setup of DreamHack then was to have to core switches, one in each event hall, and a 100G connection between the core switches. Connected to the core switches was distribution switches, each taking care of the traffic from one or more lan-tables. And each table had three access-switches.
When we ran the scripts the data started to flowing in, it was like an avalanche. But quite quickly we noticed that something was amiss. The monitoring system that made sure that all equipment was up and running started to show alarms, half the event was without internet! Quickly someone radioed into the other hall and asked a crew to check the connection there, but everything worked as it should, it was only the monitoring part that had gone down.
So, even if the scripts didn’t cause any outage for the participants, we shut down the scripts and didn’t continue with the testing.
The code
The code used seems to be lost, but I found some early version in a backup once that I uploaded to Github, you can find it here: https://github.com/the-louie/322