Dear Bernd, Really thank you for your advice. What makes me to have this thought is I want to improve the performance of my software. Normally, I use user-space socket interface to receive UPD/TCP packets. But for my project, in fact, another device will process the payload of the packets, so I have to send the data to kernel driver again. That means I have to do a lot data transfer, such as from kernel to user-space, then from user-space to kernel again. If I can directly receive TCP/UDP packets from kernel space and send them into my device. Does that mean I can save a lot of CPU time? Of course, I'm not sure if such this crazy thing can help me, maybe it is a very silly idea. In addition, I need to receive high bit rate data stream [20Mbit/s] from ip port. So performance is really important for me. Terry -----Original Message----- From: kernelnewbies-bounce@xxxxxxxxxxxx [mailto:kernelnewbies-bounce@xxxxxxxxxxxx] On Behalf Of Bernd Petrovitsch Sent: Friday, November 09, 2007 5:01 PM To: Tang Rui Cc: kernelnewbies@xxxxxxxxxxxx Subject: Re: Any one has experience about TCP/UDP develepment in LINUXkernel space. On Fri, 2007-11-09 at 10:17 +0800, Tang Rui wrote: [...] > I want to develop a TCP/UDP daemon in LINUX kernel-space, I use > this demon to receive data from specified ip port, and then injector > the data to another driver or device. > > Most of cases, we can use user-space socket interface to receive > TCP/UDP packet , then send data to driver by driver's user-space file > control interface , > > Why not we do such thing in kernel space? Dose any one have It buys much less performance then you think/believe if you implement it sanely in user-space - e.g. learn about the sendfile() and splice() system-calls. If you need almost-realtime, learn about POSIX real-time priorities. And there are other means to solve other potential problems good enough for almost all applications. If you need real hard real-time, learn about RTAI. And you have to cope with all the kernel-internal programming restrictions, non-existing debug features, risk of killing/booting/freezing the system just because your daemon has a silly bug. > experience about this? No, because in almost all cases the network traffic is slow enough for user-space. So perhaps you can describe what's the primary reason you want to so it in the kernel? Bernd -- Firmix Software GmbH http://www.firmix.at/ mobil: +43 664 4416156 fax: +43 1 7890849-55 Embedded Linux Development and Services -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
