Re: [linux-audio-user] what window manager are you using?

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: Re: [linux-audio-user] what window manager are you using?
From: Erik Steffl (steffl_AT_bigfoot.com)
Date: Fri Aug 13 2004 - 01:46:49 EEST


Lee Revell wrote:
> On Thu, 2004-08-12 at 15:41, Erik Steffl wrote:
>
>>Lee Revell wrote:
>>
>>>99.99% of Linux users should use their vendor's
>>>kernel. It has undergone a LOT more stability testing than whatever you
>>>would compile off of kernel.org.
>>>
>>>You should only use a kernel.org kernel if you need some feature or
>>>driver that your vendor's kernel does not provide, or, obviously, if you
>>>are hacking the kernel. If you go this route you should attempt to
>>>build a binary package for your distribution, then install that. This
>>>way you can post the packages somewhere, and other people who need a
>>>custom kernel for their own purposes can just download your packages vs.
>>>repeating all that work.
>>
>> I think he was asking a different Q: he was asking about advantages
>>of using compiled as opposed to pre-built kernel, you are comparing
>>distro kernel to vanilla kernel.
>>
>> IMO: in general I think it makes sense to compile kernel because you
>>get exactly what you need plus you can experiment with different setting
>>to see if you get better performance etc. I tend to use the kernel
>>source package for my distro (mostly because it can build a package that
>>can be installed, which takes care of having LILO option to boot former
>>kernel etc.)
>
> Yes, if you only have one or two machines to worry about, then by all

   that's what the question was

> means compile your own kernel. The performance increase and memory
> savings will still probably not be enough to make it worthwhile, but you
> will at least learn a lot.
>
> Once you are dealing with more than a couple of machines, then tweaking
> your kernel *really* becomes a waste of time. Red Hat has spent a *lot*
> more time tweaking their kernel than you ever could, and they are also a
> lot better at it. Also it's SO much easier to just pop a CD in a
> machine and in 30 minutes have a system that will just work, forever,

   there are different ways to handle different situations depending on
what you want, I guess if you have enough machines you might want to
create your image and just install that instead of going through
installation on all machines... (i.e. customizing the kernel is same
whether it's one machine or many identical machines)

> versus spending half a day customizing the kernel to your hardware, only
> to get no performance gain and a few KB memory savings, and have it
> crash in a month due to some bug that RedHat already fixed.

   ???

   if you compile your own kernel that does not mean you are throwing
away bugs that redhat fixed. don't mix the two issues and then use
argument against one to argue against another.

> So, I am speaking more from my sysadmin background. Most Linux audio
> users are more interested in intensively tweaking one box rather than
> having to keep 100 of them just working. The requirements are very
> different.
>
> A good compromise is compiling your own kernel from your vendor's
> sources. Once you get it working, please post .rpms or .debs somewhere
> for others to use, especially if you applied some patch to get a feature
> that's not in the stock kernel. This is an excellent way for non-coders
> to contribute a lot to the open source process.

   did you read my email? if you get debian source it creates configured
package for you. there's no point in making it available because the
whole point is to customize the kernel to your HW (which anybody can do,
just as easily)

   Building kernels for somebody else is pretty hard and is best done if
you want to make it least hobby out of it (e.g. agnula or some almost
distro like that might want to do tat). The personal kernels are just
that - kernels customized to whatever you need. What helps is posting
what config should be used and which patches were useful (for particular
hw or particular purpose) but sharing the kernels doesn't seem
reasonable - how many people have _exactly_ same hw? How many people
that can build kernels for themselves can build generaly usable kernels?

        erik


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : Fri Aug 13 2004 - 01:55:26 EEST