Re: [manet] OLSRv2 futures

[Christopher Dearlove <[e-mail]>]

The basic RREQ/RREP idea is the same in those protocols. And why I think if we were doing that - I don’t think we will - we could start from those. But the devil is in the details. And those details don’t always work. Starting with the basic RREQ and then RREP following it backwards, how do we know links are bidirectional and that’s possible? Do we need RREQ-ACKs of some form? Then two concepts are path accumulation in RREP messages, and gratuitous RREPs. Both potentially improve performance if messages are more than just very occasional. But both introduce significant security issues. You no longer know there is a link between A and B because A or B told you that, but rather because X told you that, and A told C, who told D, who … told X. If you run with a single shared authentication key across all routers, you’ve not lost much. If you have anything more sophisticated (as OLSRv2 supports) you’ve lost more. (There are things like aggregating signatures, but I don’t think standards track is ready for them. And Iknow no more than that they exist.)

> On 8 Nov 2023, at 12:56, Abdussalam Baryun <[e-mail]> wrote:
> 
> 
> 
> On Wed, Nov 8, 2023 at 2:55 PM Christopher Dearlove <[e-mail] <mailto:[e-mail]>> wrote:
>> To summarise where I think things are with regard to OLSRv2, starting from the easiest - not necessarily the most important (if anything is).
>> 
>> - A simple standards track document that adds permission to for a router to send one or more unscheduled TC messages (probably changing schedule) on learning of a new router in the network (on adding an entry to the Advertising Router Set). However, a complication is that the motivation for this is based on the next item, so how to package the two would be an issue, as there’s a standards track/informative mismatch. Probably end up repeating some material.
>> 
>> - An informative document that explains how OLSRv2 (including NHDP) can be used in ways that allow more efficient use in some scenarios. I have notes somewhere that can expand on some of this. It could be considered as rationale for many of the design decisions. A possibly non-exhaustive list of these features and use is:
>>   + The ability to work responsively, only sending messages when the situation changes. (This might need the first point above for ideal use.) Assumes a stable network, other than joiners/leavers and the changes they offer/require, and the ability to recognise local changes. (Or continue to use NHDP as usual.)
>> + The ability to change parameters dynamically. Enables features such as exponential backoff in stable circumstances, with reversion when changes occur. Similar to the above in some regards.
>> + Different parameters on different routers and different interfaces. The former is potentially useful when e.g. some routers are moving rapidly. Ideally some modelling would support this.
>> + The ability to set multiple TC interval times at different distances, implementing the concept of fisheye/HSLS routing. Requires some, not easily defined, network properties to be suitable (density largely).
>> 
>> From this point down, there needs to be real motivation to consider them. (Also true above, but I think that might exist.)
>> 
>> - A mechanism to allow a forgetful router to rejoin a network without waiting for timeout of its previous messages. Ranked lower than the above because it’s a change, and compatibility has to be considered.
>> 
>> - A mechanism to allow a router to dump its topology information to a new neighbour for fast full joining. There are lots of questions here - solicited or unsolicited, message format (and how to make 5444 compatible when it expresses the relationship between two addresses, not a characteristic of one address), in HELLO message?
>> 
>> And a long way down, and I don’t think there’s enough commitment from the WG, or even real use cases openly demonstrated:
>> 
>> - Adding reactive features (RREQ/RREP messages) to OLSRv2. In principle this wants a reactive protocol, but as so much would need re-engineering, knowledge of AODV/AODVv2/LOADng might suffice. Although the technical problems with AODVv2 remain issues to be solved here. The basic concept is routers not prepared to do what’s needed proactively (though this has more than one level) but prepared to do some or all of the work reactively (also more than one level). The full generalisation is a form of policy-based routing.
> 
> I think the AODV algorithm is same for all just few functional changes among those versions, however, DLEP can help to trigger routers to do what is needed.
> 
> AB