Re: [Xenomai-help] Xenomai latency tests on MPC8xx

[Wolfgang Grandegger <wg@domain.hid>]

fabien wrote:
> Here, there is some xenomai tests on a MPC855T, could you comment
> these results please ?
> Are they correct ?
> 
> # cat cpuinfo
> processor       : 0
> cpu             : 8xx
> clock           : 50MHz
> bus clock       : 50MHz
> revision        : 0.0 (pvr 0050 0000)
> bogomips        : 48.38
> 
> The kernel is a 2.6.23-6 (ppc branch) from DENX git repository,
> Xenomai 2.4.0, and adeo-ipipe patch :
> adeos-ipipe-2.6.23-ppc-1.6-00.patch.
> 
> I'm new in realtime system and xenomai so i have also some questions :
> First, what's the purpose of the cyclic and switchbench test ?
> How could i interpret the results of switchtest ?
> And finaly, it's seem there is something wrong with irqloop test, but what ?
> 
> First, tests without load :
> 
> # ./latency -t0 -p1000 -T30
> == Sampling period: 1000 us
> == Test mode: periodic user-mode task
> == All results in microseconds
> warming up...
> RTT|  00:00:01  (periodic user-mode task, 1000 us period, priority 99)
> RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat worst
> RTD|      75.520|      84.480|     128.640|       0|      75.520|     128.640
> RTD|      75.520|      86.080|     136.960|       0|      75.520|     136.960
> [...]
> RTS|      75.520|      86.080|     185.920|       0|    00:00:30/00:00:30
> 
> # ./latency -t1 -p1000 -T30
> == Sampling period: 1000 us
> == Test mode: in-kernel periodic task
> == All results in microseconds
> warming up...
> RTT|  00:00:01  (in-kernel periodic task, 1000 us period, priority 99)
> RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat worst
> RTD|      56.960|      66.118|      92.800|       0|      56.960|      92.800
> RTD|      56.960|      66.777|     106.240|       0|      56.960|     106.240
> RTD|      56.960|      66.037|     136.640|       0|      56.960|     136.640
> [...]
> RTS|      56.960|      66.308|     149.440|       0|    00:00:30/00:00:30
> 
> # ./latency -t2 -p1000 -T30
> == Sampling period: 1000 us
> == Test mode: in-kernel timer handler
> == All results in microseconds
> warming up...
> RTT|  00:00:01  (in-kernel timer handler, 1000 us period, priority 99)
> RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat worst
> RTD|      18.560|      20.938|      34.880|       0|      18.560|      34.880
> RTD|      18.560|      21.446|      45.760|       0|      18.560|      45.760
> [...]
> RTS|      18.560|      21.320|     118.400|       0|    00:00:30/00:00:30
> 
> # ./cyclictest -t5 -l 30000
> T: 0 (   64) P:99 I:    1000 C:   30000 Min:      97 Act:     138 Avg:
>     127 Max:     180
> T: 1 (   65) P:98 I:    1500 C:   30000 Min:      97 Act:     101 Avg:
>     195 Max:     431
> T: 2 (   66) P:97 I:    2000 C:   30000 Min:      97 Act:     138 Avg:
>     290 Max:     660
> T: 3 (   67) P:96 I:    2500 C:   25569 Min:      95 Act:     881 Avg:
>     286 Max:     881
> T: 4 (   68) P:95 I:    3000 C:   21308 Min:      96 Act:     168 Avg:
>     552 Max:    1354
> 
> # ./switchbench -n10000 -p 1000
> == Sampling period: 1000 us
> == Do not interrupt this program
> RTH|     lat min|     lat avg|     lat max|        lost
> RTD|     137.920|     143.040|     160.640|           0
> 
> # ./switchtest -T30 -n
> == Threads: sleeper-0 rtk-1 rtk-2 rtup-3 rtup-4 rtus-5 rtus-6 rtuo-7 rtuo-8
> RTT|  00:00:01
> RTH|ctx switches|-------total
> RTD|         999|         999
> RTD|         993|        1992
> RTD|        1005|        2997
> RTD|         999|        3996
> RTD|         993|        4989
> RTD|        1005|        5994
> RTD|         999|        6993
> RTD|         990|        7983
> RTD|         993|        8976
> RTD|        1005|        9981
> RTD|         999|       10980
> [...]
> 
> # insmod ./xeno_irqbench.ko
> # ./irqloop -t0
> irqloop: error starting test: Inappropriate ioctl for device
> Xenomai: POSIX: destroyed thread c03a0e10
> # lsmod
> Module                  Size  Used by    Not tainted
> xeno_irqbench           9788  -
> xeno_posix            179428  -
> xeno_switchtest         8704  -
> xeno_rtdm              30572  -
> #
> 
> 
> 
> ----
> Now tests with load : dd if=/dev/zero of=/dev/null
> ----
> 
> # ./latency -t0 -p1000 -T30
> == Sampling period: 1000 us
> == Test mode: periodic user-mode task
> == All results in microseconds
> warming up...
> RTT|  00:00:01  (periodic user-mode task, 1000 us period, priority 99)
> RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat worst
> RTD|     108.160|     119.680|     133.760|       0|     108.160|     133.760
> RTD|     115.840|     120.000|     154.240|       0|     108.160|     154.240
> [...]
> RTS|      95.040|     119.680|     202.880|       0|    00:00:30/00:00:30
> 
> # ./latency -t1 -p1000 -T30
> == Sampling period: 1000 us
> == Test mode: in-kernel periodic task
> == All results in microseconds
> warming up...
> RTT|  00:00:01  (in-kernel periodic task, 1000 us period, priority 99)
> RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat worst
> RTD|      65.920|      73.409|     100.480|       0|      65.920|     100.480
> RTD|      65.920|      74.055|     107.840|       0|      65.920|     107.840
> RTD|      65.920|      73.954|     100.800|       0|      65.920|     107.840
> [...]
> RTS|      65.920|      73.868|     132.480|       0|    00:00:30/00:00:30
> 
> # ./latency -t2 -p1000 -T30
> == Sampling period: 1000 us
> == Test mode: in-kernel timer handler
> == All results in microseconds
> warming up...
> RTT|  00:00:01  (in-kernel timer handler, 1000 us period, priority 99)
> RTH|-----lat min|-----lat avg|-----lat max|-overrun|----lat best|---lat worst
> RTD|      21.120|      25.826|      39.680|       0|      21.120|      39.680
> RTD|      21.120|      26.073|      39.360|       0|      21.120|      39.680
> [...]
> RTS|      20.800|      25.987|      54.400|       0|    00:00:30/00:00:30
> 
> # ./cyclictest -t5 -l 30000
> T: 0 (   68) P:99 I:    1000 C:   30000 Min:      96 Act:     135 Avg:
>     144 Max:     177
> T: 1 (   69) P:98 I:    1500 C:   30000 Min:      96 Act:     130 Avg:
>     216 Max:     483
> T: 2 (   70) P:97 I:    2000 C:   30000 Min:     128 Act:     133 Avg:
>     313 Max:     704
> T: 3 (   71) P:96 I:    2500 C:   30000 Min:      98 Act:     136 Avg:
>     292 Max:     932
> T: 4 (   72) P:95 I:    3000 C:   30000 Min:     127 Act:     135 Avg:
>     470 Max:    1378
> 
> # ./switchbench -n10000 -p 1000
> == Sampling period: 1000 us
> == Do not interrupt this program
> RTH|     lat min|     lat avg|     lat max|        lost
> RTD|     144.960|     156.160|     162.240|           0
> 
> # ./switchtest -T30 -n
> == Threads: sleeper-0 rtk-1 rtk-2 rtup-3 rtup-4 rtus-5 rtus-6 rtuo-7 rtuo-8
> RTT|  00:00:01
> RTH|ctx switches|-------total
> RTD|         999|         999
> RTD|         993|        1992
> RTD|        1005|        2997
> RTD|         999|        3996
> RTD|        1008|        5004
> RTD|         993|        5997
> RTD|        1005|        7002
> RTD|         999|        8001
> RTD|         999|        9000
> RTD|         999|        9999
> RTD|         993|       10992

First, it's nice to see that all tests passed without oops, system hang
or large latency figures. Your results are reasonable for your test
scenario. I remember similar results on my TQM860L module under Linux
2.4.25. Your target is a very low-end PowerPC system with little cache
(min latency is rather close to max latency) and it's already a pain to
run Linux 2.6 on it. But you should apply more load and run the test
much longer to get realistic worst case latency figures, which are
likely higher then the values listed above. I usually use "while ls; do
ls; done" in a telnet window, "ping -f <host-ip>" and the cache
calibrator to produce high load. And do not forget to add "cat
/proc/xenomai/latency" to the above figures.

Wolfgang,