Dan Williams wrote:
You may want to update to the Q2B81 firmware which disables the branch
prediction workaround, and yields about 25% performance increase when
using any USB network devices (including the builtin wireless). You
don't have to reflash the NAND to do this, so you wouldn't need to blow
away the setup you've already got.
dan
I updated the machine to Q2B81 firmware and reran the experiment with the
machine being idle. The percentage of time spent in the idle looks the same. I
do see the "percent halted" go down when I put load on the machine. The "exited
default_halt" is lower than the "entered default_halt" in this particular data run.
-Will
# staprun idle1.ko
Starting halt watch
Ran for 647276729 us
entered default_halt 100228
exited default_halt 100204
percent halted: 17
count: 100204
avg_time: 1140
min_time: 1
max_time: 3438
usec distribution
value |-------------------------------------------------- count
0 | 0
1 | 1
2 | 212
4 | 911
8 | 1070
16 |@ 1689
32 | 1386
64 | 1392
128 | 1503
256 |@ 2753
512 |@@@ 5264
1024 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ 83782
2048 | 241
4096 | 0
8192 | 0
_______________________________________________
Devel mailing list
[email protected]
http://mailman.laptop.org/mailman/listinfo/devel
