I have not even decided if I want to undertake a kernel build, but for the sake of discussion, if everything happens to be in the stock kernel, why, other than file bloat, would you want to build a custom kernel? In FreeBSD, you can remove all those devices and drivers that are not needed, and, depending on how much stuff you add or delete, can make a decent size reduction in the kernel itself. I'm not smart enough to know if there is a performance trade-off with a whole bunch of unneeded modules or drivers in the kernel, but perhaps common sense tells me that there must be some kind of performance hit otherwise. Kernel building in Linux is very much different from the BSD camps, and I'll admit I'm not sure I understand enough about the Linux build process to undertake it.
Comments appreciated......
On Thu, 2005-09-22 at 09:52, Sam Drinkard wrote:
I have not even decided if I want to undertake a kernel build, but for the sake of discussion, if everything happens to be in the stock kernel, why, other than file bloat, would you want to build a custom kernel? In FreeBSD, you can remove all those devices and drivers that are not needed, and, depending on how much stuff you add or delete, can make a decent size reduction in the kernel itself. I'm not smart enough to know if there is a performance trade-off with a whole bunch of unneeded modules or drivers in the kernel, but perhaps common sense tells me that there must be some kind of performance hit otherwise.
Nearly all drivers are built as modules which have no effect on the kernel if you don't load them. If the disk space used by the module bothers you, you can remove the unused ones. The main downside to building them all is the time it takes to compile a kernel which most people don't need to do.
Les, is there any performance hit by using the stock kernel? One reason I ask, is the application that hopefully will be running if we can get the compile done is very number-crunching intensive. In fact, that is about all it does do aside from generating a few files. Altho I've got more than adequate horsepower to handle the job, I was just curious if there were any "standard reasons" to either build or not build. As for file sizes, no problem on space. I see from the bit I've looked around at the filesystem the kernal is not too large as it is, but of course, I don't know what the vsize is at runtime either. I know there is a lot of reading material in the man pages, and other sources too, but to get some more generalized info to start with is my goal.
Les Mikesell wrote:
On Thu, 2005-09-22 at 09:52, Sam Drinkard wrote:
I have not even decided if I want to undertake a kernel build, but for the sake of discussion, if everything happens to be in the stock kernel, why, other than file bloat, would you want to build a custom kernel? In FreeBSD, you can remove all those devices and drivers that are not needed, and, depending on how much stuff you add or delete, can make a decent size reduction in the kernel itself. I'm not smart enough to know if there is a performance trade-off with a whole bunch of unneeded modules or drivers in the kernel, but perhaps common sense tells me that there must be some kind of performance hit otherwise.
Nearly all drivers are built as modules which have no effect on the kernel if you don't load them. If the disk space used by the module bothers you, you can remove the unused ones. The main downside to building them all is the time it takes to compile a kernel which most people don't need to do.
On Thu, 2005-09-22 at 10:57, Sam Drinkard wrote:
Les, is there any performance hit by using the stock kernel? One reason I ask, is the application that hopefully will be running if we can get the compile done is very number-crunching intensive. In fact, that is about all it does do aside from generating a few files. Altho I've got more than adequate horsepower to handle the job, I was just curious if there were any "standard reasons" to either build or not build.
I haven't built a custom kernel since the modular approach included everything I needed somewhere in the RedHat 6.x era. I can't think of any reason to do it other than needing to add drivers that won't work as modules.