Repository : ssh://git@open-mesh.org/doc

On branches: backup-redmine/2017-07-17,master

commit d852f81a68df212eb9b04e6ae89519af260fe5b0 Author: Sven Eckelmann sven@narfation.org Date: Sun Jul 16 21:28:22 2017 +0000

doc: batman-adv/OGM: Fix multiple page titles

d852f81a68df212eb9b04e6ae89519af260fe5b0 batman-adv/OGM.textile | 44 ++++++++++++++++++++++---------------------- 1 file changed, 22 insertions(+), 22 deletions(-)

diff --git a/batman-adv/OGM.textile b/batman-adv/OGM.textile index 86cabc31..762b0f07 100644 --- a/batman-adv/OGM.textile +++ b/batman-adv/OGM.textile @@ -222,14 +222,14 @@ h2. Appendix

----

-h1. Questions +h2. Questions

* Where to add the description of the data packet forwarding + bonding mode description? Which object should the bonding interface stuff be added to? * Should the further optimization 'Resend OGMs with flags etc. of newest OGM' already be part of the standard?

----

-h1. Notes +h2. Notes

* Section 'Receiving other nodes' OGMs' ensures the loop-freeness - any OGM having past that part and is accepted and loop-safe and can potentially be used as a (new) router * 'Router Ranking' purges routers that are outdated in terms of sequence number and selects the routers with the highest path TQ. It further updates the status of the according router the OGm was received from. @@ -240,44 +240,44 @@ h1. Notes

----

-h1. Changes between OGMs in BATMAN IV and BATMAN V +h2. Changes between OGMs in BATMAN IV and BATMAN V

-h2. Renaming +h3. Renaming

* Global TQ renamed to Path TQ * Local TQ renamed to Link TQ

-h2. Removal of Global TQ Window and gl. TQ Averaging +h3. Removal of Global TQ Window and gl. TQ Averaging

Only remember last Path TQ and its sequence number as well as the highest sequence number of an originator node received. The window size is substituted by a SEQ_DIFF_MAX (default: 5). A neighbor node is being purged from the router list if orig_node->last_seqno - router->last_seqno > SEQ_DIFF_MAX.

-h2. Removed previous sender field +h3. Removed previous sender field

-h2. Bonding Mode using NDP's link qualities +h3. Bonding Mode using NDP's link qualities

-h2. Removal of secondary interface originators +h3. Removal of secondary interface originators

Instead when receiving an OGM, always the best link quality measured by NDP will be substracted from the OGM. This shall make multiple interfaces transparent from the OGM algorithm.

-h2. Removal of PRIMARIES_FIRST_HOP + DIRECTLINK flag +h3. Removal of PRIMARIES_FIRST_HOP + DIRECTLINK flag

PRIMARIES_FIRST_HOP flag is no more needed as there will be just one primary originator a.k. originator.

-h2. Removal of bidirectional link check +h3. Removal of bidirectional link check

Asymetric Penalty should be better and enough.

-h2. (Re)Introduce strict OGM forwarding policy +h3. (Re)Introduce strict OGM forwarding policy

To avoid routing loops. MGO/batping mechanism will compensate for convergence speed performance.

-h2. Optimized Route Switching in case of outdated currently selected router +h3. Optimized Route Switching in case of outdated currently selected router

Before in BATMAN IV if a node was receiving an OGM with a sequence number that caused the currently selected router to be moved outside of the global TQ window (e.g. receiving an @@ -293,7 +293,7 @@ It will switch to the router with the highest path TQ which is still in the OGM_ and rebroadcast its buffered OGM instead of the just received OGM. The just received OGM will still be buffered (router-addr + seqno + path-TQ) though.

-h2. Strict hop-penalty +h3. Strict hop-penalty

In BATMAN V the hop penalty always decreases the OGM's path TQ (at least by 1/255: minimum selectable hop penalty: 1, path TQ always rounded downwards) @@ -307,32 +307,32 @@ Question: Or is the TTL check in the Router Ranking actually enough? Anyways, I monotonically decreasing path TQ values of an OGM upon rebroadcasts is probably also easier to prove to be loop-free. And shouldn't harm anything - therefore I'd have a better feeling with that change.

-h2. Increase NDP window size to 128 packets +h3. Increase NDP window size to 128 packets

With removing the Path TQ averaging, things will get too unstable otherwise. Should be later substituted with an EWMA [1].

----

-h1. Further Ideas for Optimizations +h2. Further Ideas for Optimizations

-h2. Resend OGMs with flags etc. of newest OGM +h3. Resend OGMs with flags etc. of newest OGM

An accepted router with the highest sequence number has the most up-to-date information about 'Flags', 'Gateway Flags' (and also TT *?), they do not have to be stored once per router, once per originator entry is enough. And any rebroadcasted OGM could update these fields from the most-up-to-date router. Might make things a little more "complex".

-h2. Also aggregate different packet types +h3. Also aggregate different packet types

For instance NDP + OGMs to reduce number of packets sent.

-h2. Positive Feedback OGM rebroadcasting +h3. Positive Feedback OGM rebroadcasting

When for any originator thes compared to the link quality used for the last rebroadcasted OGM, resend the same OGM but with the path TQ multiplied with the new, better link TQ value instead.

-h2. No OGMs if no hosts +h3. No OGMs if no hosts

To be able to reduce the overhead by just putting some intelligent "repeater" nodes somewhere without them sending their own OGMs

-h2. OGM forwarding optimizations in asymmetric neighborhoods +h3. OGM forwarding optimizations in asymmetric neighborhoods

An asymmetric link can either mean that (a) TQ >> RQ or (b) TQ << RQ. For OGMs only (b) causes trouble and reduces the propagation time of an OGM @@ -349,12 +349,12 @@ for most of its neighbors the link is asymmetric with TQ << RQ.

The scenarios could be detected via NDP.

-h3. Extra OGM Unicasting +h4. Extra OGM Unicasting

In the Church-Tower-Scenario extra OGMs could be forwarded via unicast to the few nodes. (Extra OGM broadcasts would be unfair for the other neighbors)

-h3. Extra OGM Broadcasting +h4. Extra OGM Broadcasting

For the "Valley"-Scenario an OGM could be broadcasted more than once. (Extra OGM Unicasting might result in too many packets) \ No newline at end of file