For each, in sequence the commands
make clean
time make -jn zImage
are applied for successive values of n=4,4,5,6, and 7.
Since in real life one hopes the cache to contain useful "stuff",
my idea is to simulate this with the first compile, in order
to "precondition" the cache/buffers.
Second and subsequent compiles with increasing j-values to simulate
performance under increasing cpu load. Granted it's very artificial;
just trying to mimic my typical use pattern; and also trying to avoid
case where memory starvation is major factor.
Reported are total elapsed time (minute:sec) for each compile.
Processor is K6-266 with 32MB memory, 512k cache memory.
Kernel | j4(1) j4 j5 j6 j7
--------------------------------------
220-6 | 4:53 4:40 4:53 5:15 5:54
220-5 | 4:50 4:39 4:55 5:00 6:16
220-4 | 4:59 4:45 4:57 5:34 5:52
2.1.131| 4:49 4:45 5:20 6:25 8:02
--------------------------------------
It appears the 2.2.0-pre series are making better use of the
"preconditioning", with pre5 and pre6 showing the smoothest
response to increasing j values for the 'make'.
For the qualitative look and feel, I have an informal personal
benchmark which involves rapidly displaying a cyclic sequence
(51) of large (600x800) X11 bitmaps while doing text editing
with xemacs, which I have been following since version 2.1.109.
Strong preference for the pre-6 kernel among all those tested.
~--
Robert Hamilton
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/