Re: [manet] Problems not reports (was Re: AODVv2 implementation)

[Christopher Dearlove <[e-mail]>]

Technical issues with AODV (RTF 3561) have often been discussed on many occasions. They include not including tests for bidirectionalitry of links, lack of use of link metrics, and mutable messages and intermediate RREPs meaning that authentication (not in RFC3561, but it’s going to be needed)  is hard - specifically knowing what is authenticated by whom. That’s why a standard track approach would need to build on AODV, not just adopt it. And would need implementation. (And others have pointed out some other issues.)

> On 28 Jan 2024, at 22:47, Abdussalam Baryun <[e-mail]> wrote:
> 
> On Thu, Jan 25, 2024 at 8:07 AM Donald Eastlake <[e-mail] <mailto:[e-mail]>> wrote:
>> I think that further exploration of AODVv2 and any AODV implementation problems would be useful topics for the MNET WG and that we should consider adding that to the Charter.
> 
> IMHO, there are/was no problems reported by any participant within any IETF list or any IETF WG meeting regarding AODV (RFC3561) or any MANET experimental RFC. Please refer me to the problems so we can discuss them one by one on this WG list.
> 
> However, last MANET WG meeting there was input that implementing Multicast has many problems for MANET, which the input was without details. So we will need a report in this WG about what was those Multicast problems especially for our RFC6621, to help progress.
> 
> Best Regards,
> AB
>  
>> On Mon, Dec 4, 2023 at 8:42 AM Adamson, Robert B CIV USN NRL (5592) Washington DC (USA) <brian.adamson=[e-mail] <mailto:[e-mail]>> wrote:
>>> As Justin Dean mentioned, our nrlsmf code can act as tun/tap user-space forwarding engine (and works on Linux (including Android ... although newer Android policies have locked out use of some the the netlink APIs and I haven't tested that in a while), OSX/BSD, and even Windows with use of WinTAP (or whatever it's called).  We have done some stuff with that using our own forwarding tables or with the OS tables but not yet for AODV.  For AODV some sort of extra packet caching could be useful or do something similar with what we do with the Elastic Multicast implementation to flood until the "on demand" route is established.  Related to that the "nrlsmf" code has support for some hybrid policies that can be set related to flooding/forwarding such as token bucket limited forwarding rate until a route is established or flooding a small number (e.g, burst) of packets initially upon a new flow until a route is established.  With a little tweaking, these could be extended a little bit to support other caching strategies if desired.
>>> 
>>> Also, with the code, I have _slowly_ been refactoring code to keep the forwarding plane aspects of the the nrlsmf "engine" distinct and separate from the control plane aspects with the idea in mind to eventually have the forwarding plane code be used independently in a kernel module (or elsewhere) with the various forwarding plane features we see needed for these sorts of protocols including duplicate packet detection, control plane updates like mentioned for AODV2 below, per-flow forwarding state, etc.  It's been _slow_ since my time to work on it is quite limited these days.  Meanwhile, we do get pretty good performance with it in user space, and I know there's improvements that could be made.
>>> 
>>> If the kernel forwarding table is used, there are ways (not pretty) to update by using a pcap socket to monitor packets and do that independently of the actual forwarding.  Again, not pretty or efficient, but a work around until such features are resident in the forwarding engine.  That's one of the things I see needed, including for the Elastic Multicast protocol operation, to have the forwarding plane notify the control plane, at least periodically and/or on an event-driven basis (e.g., new flow detected).  On Linux, I think that ideally would be through a netlink interface and my thought on encapsulating a forwarding mechanism like what "nrlsmf" provides into a kernel module would include that interface.  For example, to support the "per flow" packet processing and such notifications, I created a "Flow Description" data structure that's also used as an index on a modified Patricia Trie prefix-based radix tree (with field wildcarding support) to detect, identify, and classify packet flows for such purposes.
>>> 
>>> Cheers,
>>> 
>>> Brian
>>> 
>>> On 12/4/23, 9:04 AM, "manet on behalf of Henning Rogge" <[e-mail] <mailto:[e-mail]> <mailto:[e-mail] <mailto:[e-mail]>> on behalf of [e-mail] <mailto:[e-mail]> <mailto:[e-mail] <mailto:[e-mail]>>> wrote:
>>> 
>>> 
>>> Hi,
>>> 
>>> 
>>> I was thinking about how to implement AODV2 on a TUN but hit an
>>> interesting issue when reading through the "Interaction with the
>>> Forwarding Plane" chapter of AODVv2...
>>> 
>>> 
>>> Chapter 6.4:
>>> "AODVv2 needs to update the record of when a route was last used to
>>> forward a packet".
>>> 
>>> 
>>> Does this mean AODV cannot use the kernel routing forwarding system
>>> and is expected to do everything in user space? I ask because the
>>> "routing cache" (and its statistics for each route) were removed from
>>> Linux in version 3.6.
>>> 
>>> 
>>> Henning Rogge
>>> 
>>> 
>>> _______________________________________________
>>> manet mailing list
>>> [e-mail] <mailto:[e-mail]> <mailto:[e-mail] <mailto:[e-mail]>>
>>> https://www.ietf.org/mailman/listinfo/manet <https://www.ietf.org/mailman/listinfo/manet>
>>> 
>>> 
>>> 
>>> _______________________________________________
>>> manet mailing list
>>> [e-mail] <mailto:[e-mail]>
>>> https://www.ietf.org/mailman/listinfo/manet
>> _______________________________________________
>> manet mailing list
>> [e-mail] <mailto:[e-mail]>
>> https://www.ietf.org/mailman/listinfo/manet
> _______________________________________________
> manet mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/manet