[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

[i386] introducing a kernel 2.4 installation flavor



Hello Adam, hello everyone. 

As promised on IRC, I have worked on a useable solution for using the
2.4 kernels in our current boot floppies and I think I have one.

Arguments first:
we should allways trade-off between stability and
actuality of Debian components. Currently we do not know how long the
freeze will take so nobody wants to make big jumps. So I did not want to
make big changes on the current boot floppies, and decided to create a
bf2.4 flavor parallel to the current ones (except of reiser and
udma100-ext3).

Currently, I see a timing problem in the release cycles of Linux kernels
and Debian Woody. While we are trying to make Woody stable, more and
more time passes. The kernel intended for Woody on i386 (the most used
arch) is 2.2.20. Please note that this generation is tested and known to
be stable, though it is missing important features that cannot be fully
backported from 2.4.x. Also note that the main development of the 2.4
branch is stopped and only bugfixes and important updates are done
there. I do no longer expect buggy releases. Now it is becoming stable
while the old generation has more and more upcoming compatibility
problems. This, and the fact that kernel 2.4.x has been tested by other
distributions for more than one year, should show that it is ready for
Debian, even on the i386 architecture.  Don't forget that the freeze can
take some months. In this time, we make a good test of 2.4using boot
floppies.

Some pros:
 - better behaviour on modern hardware
 - better performance
 - new filesystems
 - no 1M size limit. With 2.2.x, we currently waste about 300KiB on the
   first disk due to this limitation.
   
Contras:
 - see above. Someone may say, 2.4.x is less stable. But before judjing
   this way, first check whether such statement are based on the quality
   of old kernel components or on problems in the new (optional)
   components which have never existed in 2.2.x so you naturally cannot
   face the particular problem there.

What, if we wont change to 2.4:
 - You won't be able to install Debian on modern hw. You won't be able
   to use modern NICs, modern IDE controllers (read: no harddisk
   access), etc. Though IDE patch is available for 2.2.x, it is buggy
   (reported to hang on HD detection with new chipsets) and not
   frequently updated. Same for other patches. We can add new. wheels to
   the old car, or just take the modern one.
 - You wont be able to install on modern Reiserfs format, while the old
   (in 2.2) is reported to be buggy.
 - You will cause unneded headaches for many users because of poor
   hardware detection functions in 2.4.x. For example, take a typical
   ISA souncard and a popular BT TV card. You will have to do lots of
   RTFM to configure them with 2.2.x, while the autoconfiguration works
   fine in 2.4.x so the devices are ready just after modconf'ing the
   module.
 - We would loose our good image. Kernel 2.2 is seen as absolete by most
   people todays. Imagine, Woody+1 will be released in 1.5-2 years. Then
   most distributions are using stable and prooved 2.4.22 or so, while
   this odd software museum called Debian is sticked to 2.2.something
   from middle ages.

Realisation:

I know, our official kernel-image-2.4 packages created by Herbert Xu are
using initrd and are not supposed to replace 2.2 without additional
configuration. This should not be changed, so we should create some
versions with essential drivers built-in, as done in 2.2.x packages. In
summer 2001, I have created a set of boot floppies based on 2.4.9 and
the config of kernel-image-2.2.19. I did not receive any bug reports
related to the kernel version. So I suggest to create this flavor:

 * "bf2.4" with kernel-image-2.4.17-bf2.4. This package is created by me
   with following assumptions:
   - merged from current vanilla, udma100-ext3 and reiserfs
   - have at least the same drivers as reiserfs flavor
   - be at most as big as udma100-ext3
   - with framebuffer and i18n
 
I think I managed it (*). IMHO, now we can replace the udma100-ext3 and
reiserfs flavor with the new bf2.4. Some people may dislike the removal
of this flavors, but see it possitive: 2.4 supports newer (more stable!)
drivers of the new filesystems and we can keep the installation kernels
without many patches.

So what do you think? IMHO most users will agree.

(*) The source files, kernel and a test build can be downloaded
soon from people.debian.org/~blade/bf2.4/

Gruss/Regards,
Eduard.
-- 
*** SignOff sbeyer: #debian.de (f**ken und haare waschen)
<Y_Plentyn> hm. warum forkt er vorm haare waschen?



Reply to: