WLAN with Linux SDIO: sneak preview

[Werner Almesberger]

Wolfgang Spraul wrote:
> "Above all, abandonment is the key to innovation

Heh, I found one, too. An unlikely source, but right on the mark:

"The first step to improvement, whether mental, moral, or religious,
 is to know ourselves - our weakness, errors, deficiencies, and sins,
 that, by divine grace, we may overcome and turn from them all"
-- Tryon Edwards (American Theologian, 1809-1894)

Now, with stable-tracking fixed again, we can learn a bit more about
the weaknesses and deficiencies of our stacks. I redid the measurement
with the Linux stack as well, because the missing drivers may have
affected the system.


===== Atheros SDIO stack =====

---- host => --------    => neo ---------
     *1000B/s ctx_sw  CPU     *1000B/s ctx_sw  CPU 
  1:   716.06     456   0%      712.71   11990  23%
  2:   720.67     391   0%      715.45   12064  25%
  3:   723.16     408   0%      717.59   11977  22%
  4:   720.36     436   0%      717.59   12021  24%
  5:   723.47     433   0%      719.74   12056  25%
  6:   715.45     286   0%      709.70   12079  25%
  7:   719.74     378   0%      716.06   12035  25%
  8:   727.23     489   0%      721.29   12040  24%
  9:   718.51     397   0%      713.92   12033  25%
 10:   721.60     476   0%      717.59   12086  25%
AVG:   720.62     415   0%      716.16   12038  24%
SDV:     3.35      55   0%        3.23      33   1%

---- neo => ---------    => host --------
     *1000B/s ctx_sw  CPU     *1000B/s ctx_sw  CPU 
  1:   809.32     289   5%      802.35   11584   0%
  2:   804.28     302   5%      798.92   11600   0%
  3:   802.35     292   5%      794.75   11559   0%
  4:   798.15     292   5%      792.13   11574   0%
  5:   759.15     293   5%      751.67   11621   0%
  6:   775.65     291   5%      768.89   11610   0%
  7:   793.62     295   5%      785.45   11596   0%
  8:   794.38     293   5%      788.40   11548   0%
  9:   794.38     295   5%      788.77   11608   0%
 10:   791.75     314   5%      784.35   11601   0%
AVG:   792.30     295   5%      785.57   11590   0%
SDV:    13.98       6   0%       14.26      22   0%

host => neo: min    /avg    /max      stddev   (100 samples)
               15.20/  72.09/ 293.00    44.88 ms
neo => host: min    /avg    /max      stddev   (99 samples)
               16.06/  79.42/ 316.45    50.36 ms


===== Linux SDIO stack =====

---- host => --------    => neo ---------
     *1000B/s ctx_sw  CPU     *1000B/s ctx_sw  CPU 
  1:   709.40     271   0%      704.93    5144  16%
  2:   704.04     457   0%      701.98    5664  16%
  3:   725.03     364   0%      719.74    5196  16%
  4:   734.23     371   0%      728.81    4844  15%
  5:   701.10     454   0%      699.63    5889  16%
  6:   726.60     416   0%      723.47    4892  16%
  7:   730.40     278   0%      726.60    5124  15%
  8:   737.78     368   0%      733.91    4748  15%
  9:   719.13     428   0%      715.14    5020  15%
 10:   731.35     425   0%      727.86    5020  15%
AVG:   721.91     383   0%      718.21    5154  15%
SDV:    12.30      63   0%       11.61     341   0%

---- neo => ---------    => host --------
     *1000B/s ctx_sw  CPU     *1000B/s ctx_sw  CPU 
  1:   738.76     937   6%      733.91   11639   0%
  2:   734.23     674   5%      728.49   11614   0%
  3:   735.84     978   6%      731.35   11629   0%
  4:   720.67     701   5%      716.98   11622   0%
  5:   724.09    1004   6%      718.82   11615   0%
  6:   714.23     945   5%      713.62   11653   0%
  7:   711.80     702   5%      707.00   11641   0%
  8:   718.20     859   5%      714.23   11626   0%
  9:   718.51     762   5%      712.41   11572   0%
 10:   741.70     747   6%      736.17   11610   0%
AVG:   725.80     830   5%      721.30   11622   0%
SDV:    10.31     121   0%        9.75      21   0%

host => neo: min    /avg    /max      stddev   (100 samples)
               16.70/  81.97/ 427.00    58.35 ms
neo => host: min    /avg    /max      stddev   (100 samples)
               16.04/  79.31/ 237.36    42.78 ms


===== Conclusion =====

The throughput results haven't changed. Ping latency is quite similar
with both stacks (and way too large), which makes me wonder if they
may have a common cause.

By the way, the scripts for measurement and report generation are in
http://svn.openmoko.org/developers/werner/wlan-perf/

- Werner