emstar/devel/state/TODO - view

Return to TODO CVS log

Jump to this file's LXR Page

Up to [CENS] / emstar / devel / state

File: [CENS] / emstar / devel / state / TODO (download)
Revision: 1.1, Wed Dec 15 23:26:17 2004 UTC (4 years, 11 months ago) by girod
Branch: MAIN
CVS Tags: rdd_alpha_version_1, pregeonet, acoustic-05-18-06, PRE_TOSNIC_FIX, PRE_CEILING_FIX, PRE_64BIT, MOTENIC_PRE_BUGFIX_20050415, LAURA_CALIBRATION_EXPERIMENTS, HEAD, ESS_RELEASE_3_5, ESS_RELEASE_3_4, ESS_RELEASE_3_3, ESS_RELEASE_3_2, ESS_RELEASE_3_1, ESS_RELEASE_3_0, ESS_RELEASE_2_0, ESS_CONNECTIVITY, ESS_CENTROUTE_TESTING, ESS2-CMS-V1_5_pretest, ESS2-CMS-V1_4cMergeSympathy_2, ESS2-CMS-V1_4c, ESS2-CMS-V1_4b, ESS2-CMS-V1_4a, ESS2-CMS-V1_3, ESS2-CMS-V1_2, ESS2-CMS-V1_1, ESS2-CMS-V1_0, EMSTAR_RELEASE_2_5, EMSTAR_RELEASE_2_1_BRANCH, EMSTAR_RELEASE_2_1, CYCLOPS_RELEASE_CANDIDATE_2_0, CYCLOPS_PRERELEASE_STABLE, CLUSTER_SYNC_4, CLUSTER_SYNC_3, CENTROUTE_EMSTAR_SOCKETS, BG_1_0, BANGLADESH_ARSENIC_1_2, BANGLADESH_ARSENIC_1_1, AMARSS_JR_DEPLOYMENT_6_05_07

added statistics output device to retx/libstatedev.
moved it out of clustersync
fixed a bug in retx that was deleting init nacks



known bugs
----------

* late-completed frags may not trigger a data device update  
* publish interface allows multiple entries with same key
  (ignores all but last, but writes all the data)

extensions
----------

* publication/removal of individual keys rather than complete
  blocks
* access to "full" data output (includes non-contiguous)
* support for many-to-one unicast flows to sink

performance improvements
------------------------

minor
-----

* init msg sets full mode at last minute when no data is present
* the checkpoint init message does not need to be pushed
* nack length field map 0x1F to 0xFFFF  
* init can have implied length field
* refresh might be made one byte shorter in some cases?

major
-----

clustersync
-----------
* passthru enabling aggregation
* prune redundant data elements to empty "RED" message

retx
----

* better aggregation of nacks and refreshes with data traffic
  (aggregation delay time before it's forced to go out)
  (if passthru in use, always send before aggregation delay)
* NACK overhearing
* acks to kill refresh messages? or integrate refreshing with 
  clustering algorithm?
* smarter nack adjustment -- knock out individual entries 
  from nacks, avoiding unnecessary nack replay?

CENS CVS Mailing List