Fixing Performance
------------------

1) Observations 

 The following observations are made when considering the current (17/11/2004)
 problems in gstreamer.

 - startup time.
   Most of the time is spend in reading the xml registry. Several methods exist
   for not using a registry (see testsuite/plugin/).
   Since registries are pluggable, one can also write a binary registry which
   should significantly reduce the size and the load time.

 - Performance in state changes
   State changes are pretty heavy currently because negotiation happens at
   each state change. The PAUSED->PLAYING state change in particular is
   too heavy and should be made more 'snappy'.
   
 - Performance in data passing.
   Too much checking is done in the push/pull functions. Most of these checks
   can be performed more efficiently in the plugins, like checking if the
   element is sufficiently negotiated. The discont event 'invent' hack used
   for fixing the synchronisation has to go away.
   
   We also propose a get_range based method for pads so that random-access 
   in file-based sources can happen more efficiently (typical for muxers).

 - Scheduling overhead.
   The current default opt scheduler has extensive data-structures and does
   uses fairly complicated algorithms for grouping and scheduling elements.
   This has performance impact on pipeline constructions and pipeline 
   execution.

 - Events in dataflow.
   Because events are put inside the dataflow, a typical chain function has
   to check if the GstData object is an event of a buffer. 
   

2) Proposed solutions

 - startup time.
   Volunteers to implement a binary registry?

 - The changes listed in the separate document 'threading' will greatly
   reduce the number of negotiation steps during state changes. It will
   also reduce the latency between PAUSED<->PLAYING since it basically
   involves unlocking a mutex in the sinks.

 - adding return values to push/pull will offload some checks to the plugins
   that can more efficiently and more accuratly handle most error cases.
   With the new sync model in 'threading' it is also not needed to have
   the 'invented' events in the pull function.

 - The following scheduling policies should be made possible for pads:

    1) get_range <-> pull-range-based
    2) get       <-> pull based
    3) push      <-> chain-based
    
   The scheduling policies are listed in order of preference. A pad should
   export what scheduling it supports and the core will select what 
   scheduling method to use. It is possible for a pad to support more than
   one scheduling method.

   It is possible that two elements cannot connect when they do not support
   compatible scheduling policies. We require that pull-based pads also
   support the chain based methods, at minimal, using a helper function to 
   queue a buffer and that get-based pads also support a push based implementation.

 - With the changes listed in 'threading' the scheduler will be a lot more
   simple since it does not have to keep track of groups and connected 
   elements. Starting and scheduling the pipeline will be a matter of
   starting the source/loop GstTasks and they will from there on take
   over.

 - move events out of the dataflow. Different methods will be used to
   send/receive events. This will also remove some of the checks in
   the push/pull functions. Note that plugins are still able to serialize
   buffers and events because they hold the streaming lock.

3) detailed explanation
 
  TO BE WRITTEN