RTEMS 4.5 Vs 4.6 Vs Linux : Performance analysis : Further Improvements: How?

[Eugene Denisov]

Hello sashti,

Wednesday, March 24, 2004, 1:50:25 PM, you wrote:

I wonder why Linux results so different from your previous test

                               Linux   RTEMS
        Time To Lock a Mutex =  0.12   0.30
      Time To Unlock a Mutex =  0.12   0.28
      Initialize a Semaphore =  0.31   0.99
         Destroy a Semaphore =  0.10   1.46
              Semaphore Wait =  0.20   0.24
              Semaphore Post =  0.19   0.31
       Thread Switching Time =  9.85   1.81

Perhaps your tests have big miscalculation of measured times exceeding
results in times.

ss> Hello All,

ss>      I have posted previously the comparison between
ss> RTEMS 4.5 and linux, Requested ideas for improvement
ss> of performance.  The suggestions were (1) Play with
ss> compiler flags (2) Much improvements were done after
ss> release of 4.5(i.e, use 4.6.).

ss>      I could succesfully build RTEMS-4.6 for
ss> pc686(Probably there is no missing makefile in pre-4)
ss> using updated toolchain.

ss>      The results of comparison between RTEMS-4.5,4.6
ss> and linux are as given below.

ss> ---------------------Time in Microseconds--------
ss> Time For:        RTEMS 4.6      4.5       Linux
ss> -------------------------------------------------
ss> Lock a Mutex        0.3         0.48        0.23
ss> Unlock A Mutex      0.47        0.49        0.37
ss> Create Mutex        0.83        0.90        0.08
ss> Destroy Mutex       0.58        0.66        0.10
ss> Semaphore Init      0.74        0.90        0.31
ss> Semaphore Destroy   0.62        1.78        0.13
ss> Semaphore Wait      0.37        0.37        0.32
ss> Semaphore Post      0.46        0.51        0.32
ss> Thread Switching    2.59        2.82       35.45
ss> -------------------------------------------------

ss> Details:
ss>    CPU & BSP   :  PIII 433 MHz pc686
ss>   Linux        :  Kernel 2.6.3
ss>   RTEMS        :  4.6 pre-4 and 4.5

ss> IMPROVEMENTS:
ss>     I wish to improve the performance of rtems such
ss> that it is close to/ better than linux.  Please
ss> suggest me ideas for this.

ss>     Regarding the compiler flags, I tried -mcpu=686. 
ss> But the performance was same as that for 386.  Please
ss> suggest me ideas for improving performance of rtems.

ss> With Regards
ss> Srinivasan

ss> __________________________________
ss> Do you Yahoo!?
ss> Yahoo! Finance Tax Center - File online. File on time.
ss> http://taxes.yahoo.com/filing.html

-- 
Best regards,
 Eugene                            mailto:dea at sendmail.ru