Re: HDLC frames delivered over PWs

>> You still have the problem of delivering the I-frames to the remote
>> LCCE reliably and in order. For that, this new fangled NSPs will need
>> to implement some sort of reliable mechanism to deliver the I-frames
>> and whatever other information may be needed (eg, XID contents, etc).
>> This reliability and resequencing is not be provided by L2TP as is.
>> At this point, you are looking at implementing something akin to LAPB
>> to deliver this information. Any of the following alternatives could
>> provide the required reordering and reliable delivery: RFC1663,
>> RFC1151, Q.921, V.42 LAPM, etc. or you could use a much simpler ack
>> based system.
>>   
>An option may be delivering L2TPv3 over TCP??, although it is not 
>considered in RFC3931

Using TCP to deliver L2TP penalizes all the protocols that do not need
reliable delivery of data frames. This is very inefficient for most
protocols to be considered. Among other problems, it duplicates effort
since the control channel contains its own method for reliable delivery,
and it is not scalable when multiple data streams are tunneled together.
Bottom line, TCP is not an alternative.

A much better alternative is to develop a mechanism to deliver data
frames reliably to be used only for those few protocols that need it.
This way, the protocols that do not need reliable delivery (the great
majority) do not have to be penalized with the performance penalty of
retransmissions, packet reordering, etc, etc.