On Sep 13, 2004, at 9:01 AM, Eric Rz wrote: If you're designing the interface between Layer-2 (Ethernet, Wi-Fi, what have you ...) and IP, as a rule, the right thing to do is to pass through packet loss rates in the 1-2% range to the IP layer. If the Layer-2 sees loss rates significantly above that on a regular basis, IP applications are known to not cope well, and so the right thing to do is to make the Layer-2 appear to have a 1-2% packet loss rate, by using techniques like retransmission or FEC. Modern Ethernet (what you buy new from Linksys or Netgear or Cisco in 2004) is switched, not shared. It achieves 1-2% loss rates extremely easily. So, stacks usually pass through the tiny loss rates of switched Ethernet up to the IP layer. This means that yes, occasionally you will see lost packets if you a UDP application (UDP is a thin layer on top of IP, one IP packet to the OS == one UDP packet to an app) on a local switched Ethernet. I've seen it with real hardware. Usually, the network is having a burst of traffic, and something -- probably the receiving network stack -- gives up and throws away a packet. But, its very rare -- 0.1% or less, if I had to put a number on it. But if that 0.1% was a NoteOff sent to an Hammond organ patch, you care :-). Thus, the recovery journal technology in RTP MIDI. Shared media wired Ethernet technology got us through the 80's. Which was a good thing :-). But it really is a technology for the history books now ... its really good history, its good to know about it because it was such a classic design, but its not what people mean anymore when they say "wired Ethernet". All that is left from that era is the bit-field -- the pattern of bits in the packet -- and the semantics of the bits. Modern wired Ethernet is switched Ethernet. --- John Lazzaro http://www.cs.berkeley.edu/~lazzaro lazzaro [at] cs [dot] berkeley [dot] edu ---