Re: [Raw] New Version Notification for draft-theoleyre-raw-oam-support-02.txt

[Rex Buddenberg <[e-mail]>]

Ethernet, for example, has a frame check sequence and a link-level (aka
layer 2) retransmit on error.  Back in the '80s there were some
experiments (by radio amateurs if memory serves) that found that end-
end (layer 4) error control was more efficient than per hop.
Nonetheless both exist.  
  John, you are correct -- TCP detected a missing packet but it had no
idea why the packet was missing. And the reactions (partially shutting
the send window) assumed that a missing packet was caused by congestion
(e.g. router queue overflow). That, of course, is a pretty naive, and
often wrong, conclusion. Especially in wireless.  
  If I'm reading what the OAM folks on this list are telling us is that
we need more information -- more 'why' in the missing packet event???
For that, you're gonna want to bore down into base station MAC
operations (via the MIB values).
  My intent here is to fix the wording in the draft to avoid attracting
the protocol police.



On Mon, 2020-04-13 at 15:31 +0200, Abdussalam Baryun wrote:
> 
> 
> On Mon, Apr 13, 2020 at 12:44 PM John Grant <[e-mail]> wrote:
> > On 13/04/2020 05:46, Rex Buddenberg wrote:
> > > > Thus, providing
> > > >    availability and reliability on top of the wireless
> > > > infrastructure
> > > >    requires specific layer 3 mechanisms to counteract these
> > > > bursty
> > > >    losses.
> > > Where I was raised, layer 4 was where reliability mechanisms
> > > (e.g. TCP)
> > > were to be found. TCP is very good at counteracting bursty
> > > losses.
> > >   The other place that bursty losses can be countered is at layer
> > > 1
> > > utilizing Forward Error Correction (FEC) techniques. In my
> > > experience
> > > in MF, LF, HF radio they can be quite effective, albeit
> > > bandwidth-
> > > expensive.
> >  In cellular there is also acknowledge/retransmit in the lower
> > layers; I guess it'd be layer 2 so still not layer 3. One issue the
> > mobile industry has is that TCP assumes a packet that gets through
> > at the second attempt has been delayed by congestion, and reacts
> > inappropriately.
> For RAW technology there needs to be reliable technique in L1 or L2
> so that we can call it reliable wireless. In cellular it is reliable
> techniques.
> 
> AB
> -- 
> RAW mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/raw