Re: [6gip] IP Address Mobility Project

[Alexandre Petrescu <[e-mail]>]

Le 23/02/2023 à 05:00, Sridhar Bhaskaran a écrit :
>> Are there proposals to replace GTP?
> 
> <SB> See 3GPP TR 29.892 - 
> https://www.3gpp.org/ftp/Specs/archive/29_series/29.892/29892-g00.zip

Thanks for the pointer to an authoritative GTP spec.

[...]
>> I can say that QoS can also be done with flow labels and traffic
> classes
>> in IPv6 headers, without GTP.
> <SB> Flow labels and traffic classes can be changed by on path IP 
> devices (routers etc).

Hm?

Within an xGPP operator network, the flow label must not be changed by
intermediary nodes, because there is only one authority, there are SIM
cards and link-layer security, and thus security is easily manageable
(the RFCx spec says "A forwarding node MUST either leave a non-zero flow
label value unchanged or change it only for compelling operational
security reasons as described in").

The security mentioned in that RFC, which might make people want to
modify the flow label, is the insecurity of a covert channel.

I think in a xGPP (3GPP) network it's relatively difficult to make use
of covert channels; an end user might tweak the smartphone and put
covert channel data in a GTP header (if the GTP header is visible to
UE), but the receiver P-GW where GTP ends is certainly not in the hands
of attackers.

But let me ask whether GTP resists to covert channel attacks?  I.e.
someone transmitting inconpiscuously some data in a flow identifier
field (a GTP field) that nobody looks at?

Then there are the thousand 3D firewalls within xGPP networks that look
at all fields and doubt about everything, so I am not sure about,
overall, whether there is any reason to _not_ use flow labels of IPv6
and rather put something new (flow identifier) in GTP.

> 3GPP endpoints require an unmodified flow identifier. That is why 
> they keep the flow identifier in additional layer of encapsulation 
> (i.e in GTP header).

They should use the flow label in the base header of IPv6, and not an
additional header.

> The flow identifier marked by the first 3GPP entrypoint gateway
> (UPF) is kept all the way upto radio network in the downlink
> direction so that, radio network can directly use that as index to
> look up the resource allocation scheme / scheduler and RRM details.
> For certain services like voice media (which is marked by QOS flow
> identifier corresponding to QCI/5QI 1), mission critical services
> (QFI --> 5QI --> 69) etc, strict SLAs are there. Network cant afford
> to let on path routers meddle with QoS flow markers.

It sounds reasonable.

However, I cant refrain from wondering why doesnt it use flow labels,
traffic classes, intserv, diffserv?  Sorry again if I sound mentioning
maybe old outfashioned specs.

Alex

> 
> Regards Sridhar On Thu, Feb 23, 2023 at 1:16 AM Alexandre Petrescu 
> <[e-mail] <mailto:[e-mail]>> 
> wrote:
> 
> 
> 
> Le 22/02/2023 à 19:43, Hesham ElBakoury a écrit :
> 
> I can try to find that out.  I remember some relatively recent 
> presentations at IETF about replacing GTP, or reducing the number of
>  headers.  It might have been in the (precursor?) of the DMM WG. 
> (distributed mobility mgmt).
> 
> A related question, IMHO, would be too whether there is an RFC for 
> GTP in a first place...
> 
> I can remember an Internet Draft of year 2000 about GTP - named 
> simply draft-casati-gtp - but it was not pursued at IETF.
> 
> If there were a GTP RFC, then we could think about improving it, 
> replacing it, etc.
> 
> When there is no GTP RFC, one can think about what should an ideal 
> xGPP (3GPP) network use instead.
> 
> Or maybe another way...
> 
> Alex
> 
>> Hesham
>> 
>> On Wed, Feb 22, 2023, 10:39 AM Alexandre Petrescu 
>> <[e-mail]
> <mailto:[e-mail]> 
> <mailto:[e-mail] 
> <mailto:[e-mail]>>> wrote:
>> 
>> David, Giles,
>> 
>> Le 22/02/2023 à 14:35, David Lake a écrit :
>>> This:
>>> 
>>> “…the 5G UPF terminates the GTP tunnel but also does QoS,
> charging,
>>> lawful intercept etc. ”
>>> 
>>> … is a VERY important point.  GTP enables a number of very
> low-level
>>> RF features which MNOs need to control their allocation of
> spectrum
>>> and more importantly the numerology associated with an
> application.
>>> It is also used in shared solutions such a CoMP.
>> 
>> I can agree with the importance of the point above.  I can
> agree about
>> improvements there might be necessary.
>> 
>> I can say that QoS can also be done with flow labels and
> traffic classes
>> in IPv6 headers, without GTP.
>> 
> 
>> But I agree with the importance of the topic.
>> 
>>> If we are to replace GTP (and I think there is a reason to
> do that)
>>> we have to provide the interactions to the layer 1 that
> MNOs need.
>> 
>> I agree.
>> 
>> [...] Giles said:
>>> I’m not sure that your electricity analysis is correct.
>>> 
>>> Doing e.g. an MPLS exact match lookup across a FIB with a few 
>>> thousand labels rather than an IPv6 longest-match lookup
> across a FIB
>>> with a few million prefixes, for example, may well more
> than offset
>>> the extra electricity consumption caused by the extra 4
> bytes of
>>> header.
>>> 
>>> And then of course there’s all the electricity consumed by the 
>>> control plane updates hitting the router CPUs.
>> 
>> I can agree.  But I do not take that as an invitation to ignore 
>> electricity consumption.  Rather, all points you mention
> (exact match
>> lookup consumption, ctl plane) should be measured and told
> who consumes
>> more, and then decide what to reduce.
>> 
>> I do not take your remark as giving up electricity savings in
> front of
>> too much complexity of understanding what's happening.
>> 
>> I do not take your remark as freedom to add more headers when
> it becomes
>> necessary to add more functionality.
>> 
>> [...]
>> 
>>>> More details about this performance degradation can be
> provided if
>>>> absolutely necessary, but some times trust might also
> help for a
>>>> quicker understanding towards realizing a common goal.  (the 
>>>> development of trust is another matter that we can discuss 
>>>> separately :-)
>>> 
>>> Trust takes time to earn, and I’d tend to suggest that
> arguing that
>>> it makes sense to inject a host route for a UE into OSPF
> may not be
>>> the best way to earn it.
>> 
>> Given this apparent disparate understanding about routing -
> you seem to
>> be more people to be saying the contrary of what I say - I
> give up this
>> topic discussion on host based routes.  Maybe for later for
> another G,
>> maybe never.
>> 
>> Alex
>> 
>> -- 6gip mailing list [e-mail] <mailto:[e-mail]> 
>> <mailto:[e-mail]
> <mailto:[e-mail]>>
>> https://www.ietf.org/mailman/listinfo/6gip
> <https://www.ietf.org/mailman/listinfo/6gip>
>> <https://www.ietf.org/mailman/listinfo/6gip
> <https://www.ietf.org/mailman/listinfo/6gip>>
>> 
> 
> -- 6gip mailing list [e-mail] <mailto:[e-mail]> 
> https://www.ietf.org/mailman/listinfo/6gip 
> <https://www.ietf.org/mailman/listinfo/6gip>
> 
> 
> 
> -- o__ _>  /__ (_) \(_)... Burn fat not fuel - Bike along to a 
> healthier life and cleaner world! :)
> 
> Sridhar Bhaskaran
>