Add a comment on how G-EDF splitting works

1 files changed, 12 insertions, 0 deletions

diff --git a/litmus/sched_gsn_edf_split.c b/litmus/sched_gsn_edf_split.c
index 0fe324e2c156..bcad03fd2c9b 100644
--- a/litmus/sched_gsn_edf_split.c
+++ b/litmus/sched_gsn_edf_split.c
@@ -6,6 +6,18 @@
  *
  * This plugin is a modified version of the prior GSN-EDF plugin in
  * litmus/sched_gsn_edf_split.c
+ * 
+ * Splitting an implicit-deadline job simply means splitting each job into an
+ * integral number of subjobs.  For example, a task with a period of 10 ms and
+ * a runtime of 4 ms could be re-organized as a task with a period of 5 ms and
+ * a runtime of 2 ms, with analytical benefit for bounded tardiness (ignoring
+ * overheads and assuming no critical sections).
+ *
+ * Because our analysis works with early releasing, we actually only release
+ * each job once, but move the subjob deadline back when the appropriate amount
+ * of execution has been completed.  (In the example above, a job released at
+ * time 0 would intially have a subjob deadline at time 5, but this deadline
+ * would be moved to time 10 as soon as 2 ms of execution had completed.)
  */
 
 #include <linux/spinlock.h>