Rev 1300 | Details | Compare with Previous | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 1085 | pj | 1 | ---------------------------------- |
| 2 | Jumping Balls demo |
||
| 3 | |||
| 4 | by |
||
| 5 | |||
| 6 | Paolo Gai 1999-2001 - pj@sssup.it |
||
| 7 | |||
| 8 | ---------------------------------- |
||
| 9 | |||
| 10 | This demo was created to show some S.Ha.R.K. functionalities in the course |
||
| 11 | of Informatica Industriale , University of Pavia, Italy. |
||
| 12 | |||
| 13 | The demo is composed by an application (derived by the template application |
||
| 14 | distributed on the web site) and two init files. |
||
| 15 | |||
| 16 | The demo is composed by: |
||
| 17 | |||
| 18 | MAKEFILE The makefile used to compile the application; |
||
| 19 | demo is the rule to compile the application with a CBS scheduler |
||
| 20 | demo2 is the rule to compile the application with a RR scheduler |
||
| 21 | README.TXT This file |
||
| 22 | DEMO.H Some constants used into the demo |
||
| 1300 | pj | 23 | ECP.C, ERN.C, RRP.C, RRN.C Various initfiles |
| 1085 | pj | 24 | INITFIL2.C The EDF initfile |
| 25 | INITFILE.C A makefile that cover either EDF+CBS and RR |
||
| 26 | BALL.C The Jumping balls part of the demo |
||
| 27 | DEMO.C The main() function and some other utility function |
||
| 28 | JETCTRL.C The JET part of the demo |
||
| 29 | |||
| 30 | The demo works as follows: |
||
| 31 | - It works at 640x480 16 bit colors |
||
| 32 | - on the left, there is the jumping ball arena, on the rigth there is the |
||
| 33 | statistics about the tasks into the system. |
||
| 34 | |||
| 35 | - the tasks are guaranteed using CBS and EDF. The wcet and mean execution |
||
| 36 | time on my portable after a few minutes are (us): |
||
| 37 | |||
| 38 | JetCtrl 7400 max 7500 CBS met |
||
| 39 | JetDummy 135 max 200 CBS met |
||
| 40 | JetSlide 2100 max 2100 CBS met |
||
| 41 | Balls 276 max 380 EDF wcet for hard ball,100 CBS met for soft ones |
||
| 42 | |||
| 43 | The system should go overloaded with 40 soft balls. |
||
| 44 | |||
| 45 | - The idea is the following: |
||
| 46 | - first, an edf guaranteed ball is created. |
||
| 47 | - then, create a set of soft ball using space. since their met is < than the |
||
| 48 | real met, they posticipate the deadlines. |
||
| 49 | - if they are killed, they remain for some seconds in the zombie state |
||
| 50 | - if a set of soft ball arde created (i.e., 10), and after a while all the |
||
| 51 | others are created, the bandwidth is fully used, and the task posticipate |
||
| 52 | their deadlines. the first set of tasks stops jumping when the bandwidth |
||
| 53 | is full utilized by the newest balls until all the tasks have similar |
||
| 54 | deadlines. |
||
| 55 | - Note on the left the slides that represents the relative ratio |
||
| 56 | between the tasks. |
||
| 57 | - Note that in overload conditions the EDF task is still guaranteed |
||
| 58 | - Note that PI is used with EDF, also if no theory says that it is good: |
||
| 59 | - S.Ha.R.K. allows that, it is the user that have to choose if that is |
||
| 60 | a non-sense |
||
| 61 | - PI is independent from the implemnentation of the scheduling modules |
||
| 62 | - if the second init file is used, RR is used instead of EDF+CBS. |
||
| 63 | - In overload condition RR perform differently from EDF+CBS, giving to |
||
| 64 | each task an equal fraction of bandwidth |
||
| 65 | |||
| 66 | - note also: |
||
| 67 | - the redefinition of the standard exception handler |
||
| 68 | - the redefinition of the keys and the initialization of the keyboard |
||
| 69 | - the myend exit function |
||
| 70 | - the main() that terminates |
||
| 71 | - the two parts (ball and jet) can be excluded using a #define |
||
| 72 | |||
| 73 | If You have any question, please contact the author... |
||
| 74 | |||
| 1299 | pj | 75 | Update (2003/12/19): |
| 76 | Just some more notes... |
||
| 77 | I usually show the four demos in the following order: |
||
| 78 | |||
| 79 | (first of all, tune the demo depending on how fast your notebook is; see demo.h) |
||
| 80 | |||
| 81 | ECP - EDF with CBS and PI... the white EDF ball still work also in overload; CBS balls just slow down but they still get their guaranteed bandwidth |
||
| 82 | |||
| 83 | RRP - Round Robin with Priority inheritance ... the EDF ball is at the same level of the others, all slow down... |
||
| 84 | |||
| 85 | RRN - 2 levels of Round Robin, without Priority inheritance ... when |
||
| 86 | there is overload, the white ball works ok, but the other have an impredicible |
||
| 87 | behavior (they execute many instances in one RR period) |
||
| 88 | |||
| 89 | ERN - EDF, RR, no priority inheritance - the white ball misses his deadline |
||
| 90 | just when there is an overload (all the other balls are queued on the same |
||
| 91 | semaphore, so the blocking time increase, and when there is overload the EDF |
||
| 92 | task misses also if it has not consumed his bandwidth). Note that it happens |
||
| 93 | also if the EDF task consume just a few microseconds... Real time does not |
||
| 94 | means "fast" :-) |