Global SD_task_schedulev (SD_task_t task, int count, const SD_workstation_t *list)

We should create tasks kind for the following categories:

Point to point communication (done)
Sequential computation (done)
group communication (redistribution, several kinds)
parallel tasks with no internal communication (one kind per speedup model such as amdal)
idem+ internal communication. Task type not enough since we cannot store comm cost alongside to comp one)

Global xbt_cfg_set_parse (xbt_cfg_t cfg, const char *options)

This is a crude manual parser, it should be a proper lexer.

Logging performance *may* be improved further by improving the message propagation from appender to appender in the category tree.

Such a datastructure was necessary/useful to store the GRAS type descriptions, but it should be reworked to become generic.