TGFF Extension Suggestions

After presenting TGFF at the IEEE CODES/CASHE'98 conference, several requests for extensions have already arisen. Later, other extensions were suggested. The following summarizes the requests for improvements that have not yet been made; comments related to these are included.

  1. Indirection with respect to tasks. Task attributes are generated directly as numbers; this essentially implies that each task reside in task-graph(s) is an independent entity. Some researchers would like generate task types/information and then to populate the task-graphs with references to these. Comment: This has been cleanly implemented in TGFF v1.0.1 and higher.
  2. Arc data. Right now, "data size" is the only attribute that can be placed on arcs, although its mean and range can be controlled. Requests for more general handling, as for tasks have been made. Comment: This has been implemented in TGFF v2.0 and higher.
  3. Multi-rate support. Support for generating multi-rate task-graphs (independent tasks accepting data via queues or instantaneously from each other) has been requested. This can be done via "creative use" of task-attributes but it is not very clean. Comment: This has been implemented in TGFF v2.0 and higher.
  4. Hierarchical task graphs. Request for task-graphs, where each task may be a task graph, has been made. This can be done by "interpreting" TGFF output now. For example, on may assume that task-graph 1 is represents the first task in task-graph 0 and that task-graph 2 represents the second task in task-graph 0; more than a single layer of hierarchy can be done using this interpretive technique as well.

General suggestions: The source code is publicly available so people may make changes on their own. We would like to make TGFF useful for as many people as possible, so please contact us if you have made a useful change so that we can include it in a future release.

Start page Packed schedules Version notes Download the source code Download the paper

Page maintained by Robert Dick.