Commit graph

60 commits

Author SHA1 Message Date
adamdunkels 37adf130b4 Rewrite of the congestion logic. A node now report it being congested
when half the transmission queue is full. When a neighbor is heard to
be congested, the rtmetric of that node is inflated to avoid sending
more traffic in its direction.

Fixed problems with beacon intervals: intervals are now not reduced
when a path gets significantly worse, only when they get better.
2010-10-11 23:42:01 +00:00
adamdunkels 0cabe483c3 Don't call sink recv function for zero-size packets, since these most likely are link estimate probes or keepalive packets. Explicitly set eseqno to zero when opening a new connection. 2010-10-07 15:52:28 +00:00
adamdunkels ad35c54473 Bugfix: when a node's rtmetric was lower than SIGNIFICANT_RTMETRIC_PARENT_CHANGE, its beacon timers would always be bumped to its lowest value 2010-10-05 20:06:47 +00:00
adamdunkels 52e9a6f1fb Turned the DRAW_TREE debug flag off by default 2010-10-04 22:36:47 +00:00
adamdunkels 65d640ac53 Several bugfixes to the proactive link estimate maintenance code, which is now enabled by default. Simplified DRAW_TREE code for drawing routing graph in Cooja. Added routing metric penalty to parents that are congested or that drop packets. 2010-10-04 20:27:50 +00:00
adamdunkels 605392522f Bugfixes and improvements: rtmetrics are now 16-bit clean. Added experimental feature that proactively probes neighbors that have not yet been tried out, in order to get an initial ETX for them. 2010-10-03 20:08:44 +00:00
adamdunkels 32ba90ea4c Bugfix: make sure that neighbor exists before attempting to update tx
count.
2010-09-28 06:53:02 +00:00
adamdunkels eed22a6f3c A number of changes to the collect code:
* Added an optional "keep alive" mechanism whereby an idle network is
periodically probed by dummy packets to maintain a recent quality
metric when there is no traffic.
* Bugfix in when new routing metrics should be advertised
* Rewrote the ACK logic so that a queuebuf is not allocated for each
ack, only for those acks generated by the sink node.
* Updated the wrap-around logic for sequence numbers: when a sequence
number wraps, it won't go back to 0. Instead, it wraps to 128. This
allows us to understand when a node has rebooted: when its seqno is <
128, it has recently rebooted.
2010-09-22 22:08:08 +00:00
adamdunkels 18e31f33b9 Minor tweaks. Removed debug output from CVS. 2010-09-14 06:48:36 +00:00
adamdunkels 591766e663 Significant improvements to the Collect protocol:
* Loop detection and untangling. Now uses in-band loop detection that
  works.

* Link quality estimation. Now starts with a conservative /
  pessimistic estimate for new links.

* Retransmissions. MAC layer and network layer retransmissions are now
  decoupled.

* Loss handling. Deals better with losses now.

* Concurrent applications. Each connection now maintains its own
  neighbor table with separated cost to sink estimates for each
  connection.
2010-09-13 13:28:14 +00:00
adamdunkels 4e0744954a Rewrite of important parts of the Collect logic: link estimation
calculation and parent selection.  This fixes a severe bug that caused
loops when packets were erroneously retransmitted due to ACK loss.
2010-09-08 19:21:45 +00:00
adamdunkels 3fa8ffda1a Moved the modules packetbuf, queuebuf, and packetqueue from net/rime to net/, since they are not Rime-specific 2010-06-14 19:19:16 +00:00
adamdunkels 2647f879c2 Bugfix: select the best parent if current parent is NULL 2010-06-14 06:54:47 +00:00
adamdunkels 3d10dedecf Added a safety measure against a problem where a node in rare situations would turn itself into a sink. Not entirely sure why this happens, but the safety measure should at least avoid the results of the problem. Added more debug messages. 2010-04-30 07:33:51 +00:00
fros4943 59357e8b75 PRINTF():ed debugging output 2010-04-01 11:08:36 +00:00
adamdunkels d7457bc340 Improved parent management: parent neighbor is retained until a better neighbor is found (with 1 ETX less than parent). Don't drop parent neighbors when loop is detected (loops are transient anyway). 2010-04-01 10:03:19 +00:00
adamdunkels 4c3b69dc07 Configuration parameter for number of MAC-layer retransmissions 2010-03-29 21:52:25 +00:00
nifi 9508d2c4d7 * Added configurable default number of transmissions for CSMA to use when not specified using packetbuf attributes (CSMA_CONF_MAX_MAC_TRANSMISSIONS).
* Renamed packetbuf attribute PACKETBUF_ATTR_MAX_MAC_REXMIT to PACKETBUF_ATTR_MAX_MAC_TRANSMISSIONS where value 0 (attribute not set) means that default number of transmissions should be used.
2010-03-26 12:29:29 +00:00
adamdunkels a4f9e62295 Added a parameter to collect_open() that specifies whether or not the node should be a router for packets, or only an originator of packets. Updated the collect module to the new annoucement API. 2010-03-25 08:51:07 +00:00
adamdunkels 7b0606cd58 Renamed the neighbor module to collect-neighbor, as it is only used by the collect module and is not a generic neighbor module 2010-03-19 13:17:00 +00:00
adamdunkels 36b995cc9a Reenabled announcement support 2010-03-16 18:12:34 +00:00
dak664 0236c6e725 Leave compiler warning for 16 bit timers 2010-03-12 17:51:07 +00:00
dak664 52c9f4bde3 Remove compiler warning 2010-03-12 15:21:14 +00:00
adamdunkels 5a6fc2ade0 Bugfix: set up rtmetric before starting to announce it. Set the number of MAC layer retransmissions correctly 2010-03-09 13:21:28 +00:00
dak664 c65e09ceb2 Remove compiler warnings 2010-03-02 15:41:23 +00:00
adamdunkels ec928a2a19 Make max queueing time depend on retransmission timeout 2010-02-28 14:15:58 +00:00
adamdunkels 7165a3866f Significant rework of the Contiki data collection protocol:
* the new version makes use of MAC-layer feedback so that bad paths
  can be identified quicker and then avoided.

* the new code uses transport layer ACKs that contain feedback from
  the collect protocol: when a packet cannot be forwarded due to lack
  of resources, the ACK contains a flag that indicates that the packet
  could not be forwarded. ACKs also contain the routing metric of the
  sender, which improves agility in face of rapid path changes.

* loop detection and management has been improved: with higher path
  metric agility, the system is more prone to short-lived routing
  loops. Instead of dropping looping packets, the new version adjusts
  the routing metric for the routes that exhibit loops so that the
  risk for future loops is reduced.

* make use of packet attributes to inform the MAC layer of how many
  times packets should be retransmitted.
2010-02-28 09:18:01 +00:00
adamdunkels 0eade2886c Better loop handling: when a loop is detected, the neighbor to which the packet was originally sent gets an inflated ETX; made sure that collect without announcments work; increased timeouts, to work nicer in larger networks 2010-02-23 18:35:23 +00:00
adamdunkels 4a522c96dd Added debugging statements for packet drops 2010-02-08 21:59:49 +00:00
adamdunkels d7b7b05d0e Moved the neigbor_init() call to the collect module, since it is the only one that currently uses it 2010-02-03 20:38:33 +00:00
adamdunkels f66cd6d44e Increased the forwarding queue to six packets to improve collection robustness 2010-02-02 21:44:05 +00:00
adamdunkels 56218aba3b Made rimeaddr_t pointers const where they should be. 2009-11-08 19:40:16 +00:00
nvt-se 33909cc2fd made neighbor discovery method configurable. default is announcements. 2009-05-30 19:54:05 +00:00
zhitao 48d4800f08 removed reference to undefined function in mspgcc 2009-05-06 15:34:28 +00:00
adamdunkels f5a57a4e54 Fixes by Ward Van Heddeghem (reapplied from versioin 1.24) 2009-04-30 20:21:06 +00:00
adamdunkels e70cf0c7c6 Rewrite of the collect module. The new version uses a packet queue for outgoing and forwarded packets so that more than one packet can be forwarded at the same time. The new version uses the packetqueue module to maintain the packet queue. 2009-04-29 20:48:57 +00:00
adamdunkels f5f52ffd1f Fixes by Ward Van Heddeghem (VUB) 2009-04-07 13:06:03 +00:00
adamdunkels 65eb5fd4e8 Renamed the rimebuf module to packetbuf to signal that the module is used outside of a pure Rime context (e.g., the sicslowpan code uses it). 2009-03-12 21:58:20 +00:00
zhitao f304941ab4 raise busy forwarding flag before sending 2009-03-10 14:07:41 +00:00
zhitao 1eb085834f Update announcement value after route metric saturates. Not doing this results in stubborn forwards from former children nodes. 2009-03-09 12:54:27 +00:00
zhitao 0a12098bd5 avoid ping pong packets during route disruption 2009-03-06 17:33:59 +00:00
zhitao 08b2315f41 bugfix: put up busy forwarding flag only after next hop found 2009-03-04 08:59:32 +00:00
adamdunkels 36bb3ece64 If no neighbor is found, we listen for announcements. Cleaned up debug output. 2009-02-16 01:18:53 +00:00
adamdunkels e1ef13dad6 Reduced the size of the packet cache from four to two to avoid falseley dismissing fresh packets (the header field that hold sequence numbers is only two bits wide, hence all sequence numbers fits in a four-entry cache...) 2009-02-10 00:44:50 +00:00
adamdunkels 3670ef2f44 Rewrote the collect module so that it uses the announcement module rather than the neighbor-discovery module. This makes it possible to implement other annoumcement back-ends than the traditional broadcast-based neighbor discovery 2009-02-09 20:58:25 +00:00
oliverschmidt 7cf3165155 Replaced NETSIM macro with general (and now actually working) solution in Makefile.include. 2008-11-17 22:52:10 +00:00
adamdunkels 31509ad726 Fixed a bunch of bugs in the collection neighbor announcement/discovery code: neighbor announcments were not repeated unless there was a change in the ETX val. Neighbor announcements happened to often when there were changes in the ETX. 2008-08-15 19:00:38 +00:00
adamdunkels ec22f3c2fc Renamed Rime nf module to netflood 2008-07-03 22:02:09 +00:00
adamdunkels c2c3835110 Changed name of Rime module ruc to unicast 2008-07-03 21:52:25 +00:00
adamdunkels 020a0c30aa Rewrote the collect module to use packet attributes instead of explicit headers 2008-06-30 09:15:22 +00:00