Re: [BENCHMARK] 2.5.59{-mm2} with contest

[Andrew Morton]

Con Kolivas <[e-mail]> wrote:
>
> io_load:
> Kernel     [runs]       Time    CPU%    Loads   LCPU%   Ratio
> 2.5.58          3       136     58.8    6       12.4    1.84
> 2.5.58-mm1      3       138     55.8    7       13.0    1.86
> 2.5.59          4       113     68.1    4       9.7     1.53
> 2.5.59-mm2      3       563     12.8    38      17.4    7.61

I don't see such gross variations here, and there's nothing between -mm1 and
-mm2 which could cause this.  (BTW: what idiot broke diffstat as distributed
in rh8.0?  It fails to understand interdiff output, which is quite
legitimately formatted).

However there are some interesting snippets.  Elapsed time for io_load:

2.5.59: 					182,183,180
2.5.59 + deadline-np-42:			198,114,106,112,183,140
2.5.59 + deadline-np-42 + deadline-np-43:	224,224
2.5.59-mm2:					212,239

So it looks like deadline-np-42.patch lessens starvation of reads by writes,
and deadline-np-43.patch considerably worsens it.

But this is all just fiddling around.  Nick is working on an implementation
of anticipatory scheduling, which is a whole new ball game.  The pressure is
on ;)