Dear Guru
First time in this list *and* kernel programing myself (I've already
followed other people code).
Some of you may have heard about "hidden node" wireless network problem.
Well, for those who haven't this problem shows itself like a collision
on a ethernet. I wrote a driver to be used to implement a polling method
over those networks (yes, it can be used to "democratize" ethernets too -
let's say it transforms a hub in a "switch" in the sense of a collision).
I input myself the job of making it compile under kernels 2.0 to 2.4.
I've tested it under 2.2 and 2.4 (the machine running 2.0 kernel is hard to
reach. I left it to a later time).
I found that under kernel 2.2 the load average is always near zero but under
kernel 2.4 it is 1. Not only that; this ksoftirqd_CPU0 proccess appeared
"eating" the CPU resources leading to a 0 idle time ("top" reading). Even with
this behavior I notice no performance penalties under both kernel series.
(to elucidate things a little: my driver launches itself over and over
under the tqimmediate queue. This is because I need milisecond precision
between function calls). I think ksoftirqd does the same keventd do.
Questions:
1) Are my observations correct?
2) Does this ksoftirqd daemon relinquish time to other applications? (this
would explain the unnoticed lost of performance. Like running seti@home under
nice -19!)
3) Some of you gurus could help me to make it a better product?
Ethy H. Brito /"\
InterNexo Ltda. \ / CAMPANHA DA FITA ASCII - CONTRA MAIL HTML
+55 (12) 3941-6860 X ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL
S.J.Campos - Brasil / \
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
IRC Channel: irc.openprojects.net / #kernelnewbies
Web Page: http://www.kernelnewbies.org/
