Re: [conex] WGLC for draft-ietf-conex-abstract-mech-06.txt

[John Leslie <[e-mail]>]

Ingemar Johansson S <[e-mail]> wrote:
> 
> One comment though on page 5, 3rd para.
> " the flow-state required for audit creates itself as it detects new
>  flows.  Therefore a flow will not fail if it is re-routed away from
>  the audit box currently holding its flow-state. "
> 
> If I map ConEx to a 3GPP LTE use case, the audit functions are best
> placed in the eNodeB. When a new (long lived) flow is created it is
> preferably preloaded with credit marks which are stored in the auditor
> in the eNodeB which the terminal is "connected" to. When the terminal
> hands over to another eNodeB the credit marks will be lost. This means
> that the ConEx markings will in this case be at least one RTT behind
> with a higher risk of false positives in the auditor in the new eNodeB
> To avoid this the credit marks would need to be forwarded to the new
> eNodeB via e.g the X2 interface.
> 
> So my question is, is it needed to add a statement that mentions this ?

   I think it would be very helpful to work out wording for this -- in
network-layer terminology. (Long-Term-Evolution alphabet-soup is beyond
what we should expect of our readers.)

   If I understand you (not a sure thing, alas!), you're saying that
when a network-layer node has reason to know the path forwarding is
changing, it should insert extra ConEx Credit marks in order that an
Audit function in the new forwarding path not interpret congestion
it observes in the presence of ConEx-Not-Marked signals to be
understatement of congestion, which calls for "some sanction" to be
applied.

   (From Section 5.5.1, "if the balance ever goes negative, the audit
function can immediately start punishing a flow, without any grace
period.")

--
John Leslie <[e-mail]>