Hello Mulyadi!
The preemption is off, because we have a Kerrighed (kerrighed.org)
patched kernel and I believe it doesn't work correctly with preemption
(at present) but I'll try that.
Ahh...welcome to clustering brotherhood :)
Oh Thanks! :-)))
I still haven't tried to change the preemption model but I'll do that
the next days because we had weekend and the short time I spent for my
work I used to understand more about the callbacks in the linux kernel
code...
I could replace the kernel thread for receiving UDP messages by an
incomplete callback solution and now I have PingPong times of ~30µsec
(over the native Ethernet interfaces I get ~23µsec) so that's an
increase of only about 4µsec in one way instead of the 1msec (with the
kernel thread solution)...
I think that I'm on the right way but there must be done more work. For
example the checksum and (I believe) also the fragmentation stuff. So I
think I must read much more kernel code and ask more people if they can
help me...
How can I read the received data from the socket, what should I do after
reading, what must be locked and when...?
And what has to be done if I want send through a socket without the
sock_sendmsg() function which can't be called in interrupt context?
Do you know any sources (books or something else) where I can read more
about socket callbacks? Because I've looked into the kernel sources
where these callbacks will be set but there are completely no comments...
>
What if you better buy a copy of Understanding Linux Network stack? I
hope that book can help you ....
That is a real good idea!
We have this book at our chair and unfortunately I could read nothing
about the socket callbacks but there is much about the sk_buff struct
and some about the checksum stuff.
Maybe that can help me with the further work...
Thanks again very much for your help and have a nice week!
Regards,
Lukas
--
To unsubscribe from this list: send an email with
"unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx
Please read the FAQ at http://kernelnewbies.org/FAQ
