Not really a problem with UDP itself, but with some very old protocols like DNS that rely on UDP but can’t be changed because of compatibility. If you’re writing a new service that uses UDP, there’s nothing stopping you from designing it so that it doesn’t provide an opportunity for bandwidth amplification.
TCP is also deciding to ramp up the amount of boxes you kick over until the post worker gets crushed by boxes, at which point you decide to lower your box-kicking rate by half and try again.
Well yes, internally that’s what it does, but from a user perspective it just looks like being handed the package, you never see any of the failed attempts (unless delivery fails completely because the company went out of business). It’s sorta more like having a butler who orders it for you and deals with any potential BS that might happen, and then just hands you the package when it finally arrives in one piece.
I mean I've been trying to formally request that ISO change the C API for send() to yeet() for sockets where connection reliability is not required at the network interface level.
UDP seems more like a ball fired from canon to me. You may not be prepared for it and you won’t know what state it’s in when it gets here, but that packet is making it to the gate no matter what. Or, in the rare case it doesn’t, it means someone else is having a real bad time.
Switches will send packets to all interfaces when using broadcasts or under extreme conditions (full MAC Address Table). This can lead to duplication if there is a loop between two or more switches and if the Spanning Tree Protocol is not used. So the answer is rarely.
That is extraordinarily rare and I’m not even sure if it’s possible anymore. That was potential attack vector in the 90’s where you have a port on network switch, and then you flood the cam table with thousands of bogus mac addresses until you fill it up, then the switch turns into a hub, and you can now sniff all traffic traversing the switch. These days I’m not sure what will happen if you do successfully fill up a switches cam table. Also cam table sizes are are much much larger now. ~128k entry’s vs maybe 1000 back in the day.
You can bring a surprisingly large number of network segments down just by plugging both ends of the same cable into a dumb switch. It probably won’t happen immediately, but eventually you will get a broadcast storm which will propagate until it hits an element smart enough to snuff it out.
Yep. That happened once. The user plugged the cable for their laptop, from the dumb switch, into the same dumb switch and took out most of the network.
Fun story. I was tasked with figuring out a connection problem on a client’s network. STP was enabled, but everyone having problems were all connected to one switch.
Some investigation later and STP’s root port is not the expected root port…
After some investigation, a user took the ethernet cable for their computer (Daisy chained off their VoIP phone), and decided to store it, in the wall jack… Across the office.
That was Jack was on a different switch, and it had a lower port cost than the primary root port between the switches, so naturally, let’s send all inter-switch traffic over to this… Telephone.
Add comment