Case in point:
One of my clients purchased a recorder for their VoIP phone system. These recorders work by capturing traffic on a line, filtering out the non-voice traffic, rebuilding the conversation, and then committing the audio and call information to storage. This particular system (whose unfortunate owner and witless provider shall remain nameless) was sold under the oft-abused title of "Plug-N-Play". Even the clients current infrastructure was evaluated and approved to be compatible by the provider. Needless to say, this was not the case.
The essential problem with the system is that the protocol used by the phones transmits audio information directly from phone to phone (not through the 3COM NBX) when you make an interoffice call. Since the clients switch is only capable of doing one-to-one port mirroring, we can only monitor one port at a time. I selected the NBX port and we were able to successfully record out of office calls, but interoffice messages were still silent.
Although I was 99% sure we simply needed a switch capable of many-to-one port mirroring, I wasn't about to recommend an expensive purchase to a client without some sort of guarantee or official recommendation. After several days of trying to get a clear answer from the provider via email (based in the UK, they have no +1 country code number and the times zones don't really match up anyways), I got sent this little dandy of a message:
As per your latest email, engineering recommends that the best way to fix this is to put all of the phones through a hub (unswitched) and place the uplink into the NBX
Yup. They want me to take dozens of phones and plug them all together on an unswitched hub, complete with collisions. For those not familiar with hubs, they are essentially the senile granddaddy's of switches. A true hub will blindly regenerate the signals from one port and repeat them out ever other port. While this would technically allow us to capture all traffic, it also puts all the phones into a single collision domain and forces them to share bandwidth. So when two phones try to send audio at the same (as is often the case in a busy office), the audio frames stand a good chance of having a collision of James Dean proportions. This is why CSMA/CD was developed, which postpones sending another frame when a collision is detected. Just dandy when you are sending, say, an email that can be delayed a half a second. Not so great when you need to transmit time-sensitive information, like a freaking phone call.