>John Neiberger wrote,
>Someone who knows more about the specifics than I do will correct me if
>I'm wrong, but if I remember correctly HDLC will not retransmit due to a
>line error. And again, IIRC, neither does PPP, frame relay, or
>ethernet. My impression is that those protocols utilize error
>detection, but not error correction. I have absolutely zero experience
>with x.25. Does it retransmit due to line errors by default or does
>that feature need to be configured?
>
>From what others have been saying, it sounds like current reasoning
>suggests that it's better if the hosts are aware of network problems so
>that upper-layer protocols can make the necessary adjustments.
It depends. There definitely are cases where combinations of slow
transmission speed, long propagation delay, and high error rates make
link-level retransmission more appropriate for optimized throughput.
Certain applications, such as voice, are more intolerant to delay (as
might be caused by retransmission) than to error. They have no error
correction whatsoever, although they have error detection that causes
them to drop errored packets.
There are other cases where forward error correction (FEC) makes
sense. FEC involves sending additional error-detecting and
-correcting bits with a frame, increasing the overhead, but allowing
the receiver to figure out what the transmitted bits were without the
need for retransmission. FEC can get quite mathematically complex,
but it is useful in certain applications where retransmission
(anywhere) would be VERY painful. Consider the extreme case, for
example, of telemetry to deep space probes where speed-of-light delay
can be in minutes or hours (Voyager? You out there?). Additional FEC
applications are found in wireless transmission, and in certain modem
applications at the bleeding edge of bandwidth for a medium.
Another variant of retransmission is SSCOP, the data link protocol
for SS7. SSCOP allows redundant links to be set up, with the
structure that if either, but not both links, receives a packet with
a bad frame check sequence, the packet is accepted only from the link
with the good FCS. Retransmission takes place only if both links
detect an error, or one link fails. This is NOT an inverse
multiplexing protocol intended to deliver twice the bandwidth over
paired links; it is intended for situations where the traffic MUST
get through and the delay of any sort of retransmission is
undesirable.
Other applications resend the data, but in a less anal-retentive
manner than SSCOP. Some digital weather facsimile broadcasts simply
retransmit the same weather map several times. Experience has shown
that in the space of 6-10 minutes, every receiver will get an
error-free copy, which is quite fast enough to get new weather
information by the time anyone can do anything about it.
There may be retransmission above the transport layer, as with
NFS/RPC. In such cases, there's no real need for the lower layers to
retransmit.
_________________________________
FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html
Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
