B.A.T.M.A.N. V outlook

Added by Marek Lindner over 14 years ago

Now that B.A.T.M.A.N. IV is out there and works quite well, it is time to merge our gained experience and work towards B.A.T.M.A.N. V. This page contains the first collection of ideas of what the improvements could look like. To keep the list size in a digestable format we had to draw a line somewhere. Therefore it only contains items which come with a plausible concept or (even better) a proof-of-concept patch. Nevertheless, it is not final yet. Over time some ideas might proove to not work or new ideas might arise. Feel free to contact us] if you want to [[Contribute.

Convergence speed:

In order to increase the convegence speed B.A.T.M.A.N. IV forwards all valid OGMs coming by and thus increasing the probability of more packets coming through. To avoid routing problems it fixes the TQ and TLL values of packets not coming from the best neighbor. It turned out this behaviour of forward and replace may cause transient routing loops under certain conditions and the convergence speed leaves room for improvements.
B.A.T.M.A.N. V will forward packets from the best neighbor only without replacing any values. Furthermore, in case of packet loss it will detect when an OGM packet from its best neighbor did not arrive and will then broadcast the missing packet by itself (using the last known sequence number & TQ value). That way, B.A.T.M.A.N. increases the probability of transmitting OGMs over a longer chain of nodes. B.A.T.M.A.N. V could send the originator interval along with the OGM to let other nodes determine whether an OGM was missed or not.

Overhead reduction:

The traffic B.A.T.M.A.N. IV generates to discover the best path through the network is quite low compared to other protocols, but especially when B.A.T.M.A.N. has many single hop neighbors which rebroadcast each others OGMs we see room for improvements. B.A.T.M.A.N. V will split the well-known originator message into two different message types. The OGM will remain but only be used to flood the TQ throughout the network. The new message type (a name needs to be found) will contain the link qualities of the single hop neighbors only. This message won't be rebroadcasted and just reaches the local neighborhood. These local message can be sent much more often than the global TQ messages and thus reduce the traffic [nearly just create a linear growth of traffic with more nodes in the local neighborhood instead of a squared amount].

Starving mode:

Fast moving nodes always have the problem of adjusting their routing information in time. They can choose to send more routing information, so that their environment can adjust to them but stationary nodes won't do the same and increase the mobile node's adaption time greatly. However, when a B.A.T.M.A.N. V node detects that his local environment changed quickly he will enter the starvation mode. In this mode the node will actively try to confirm a working route as fast as possible by sending a "batman ping" to its new neighbors. Each B.A.T.M.A.N. neighbor will try to forward the message to its destination, once arrived there it will travel back. If the mobile node receives the reply it can change its route towards the new neighbor without waiting for OGMs as the route has been verified.

Mesh bonding mode:

In certain setups it might be desirable to use 2 or more dedicated wifi links (towards your direct neighbor) at the same time to increase the bandwidth. B.A.T.M.A.N. V will provide the mechanisms to automatically detect neighbors it could bond with and establishes the bonded connection.

Incoming interface based routing:

B.A.T.M.A.N. IV will find the best route through the network which does not fully utilize the capabilities of "multi-link nodes". If a single node is able to maintain several connections via its various interfaces to its neighbors the protocol should consider switching the interfaces in order to reduce latency and increase bandwidth. To achieve this B.A.T.M.A.N. V will create a routing table on a per interface basis. Whenever a payload packet arrives at the node it will lookup the path using the routing table for that interface. Imagine a network full of nodes having 2 wifi interfaces - B.A.T.M.A.N. V would try to make use of all existing interfaces to avoid interference and maximize throughput.

There are numerous other ideas such as faster HNA handover, optimized multicast support, etc which need to mature before they can be added to the list.