Alpha Newbie needs help netbooting linux

[Mike A. Harris]

On Thu, 1 Apr 2004, Joe Brazeal wrote:

>Did anyone see the move by HP and Novell in Info Week(March 29,2004)?

No, got a URL?

>Very interesting...seems like Novell may be on the right track
>while RHN shot themselves in the foot.  In short, I would
>challenge if you took the alpha investment in terms of technical
>experience, stability, reliability, and investment and stacked
>it up against RH's decision for dropping Sun and Alpha then I
>would venture those making the decision never felt the real
>pulse...had they then Novell would have looked else where ...
>Novell and HP seem to be re-evaluating the possibilities.  
>Novell and HP may have re-discovered riches right in their own
>back yard!

Details?

>btw, "I ain't got no dawg in this here fight...just don't make
>common sense what RHN did...dropping Alpha." That is my two
>cents worth.

Not sure what you mean about "RHN dropping Alpha".  Do you mean
"Red Hat Inc."?  Red Hat produced Red Hat Linux for the Alpha
processor I believe for the first few releases and it was not
profitable for the engineering investment expended.  Later
versions of Red Hat Linux for the Alpha architecture were
produced under paid contract, and supported to the extent of the
terms of the contract.

The last Alpha product to be contracted was Red Hat Linux 7.2, 
and so that was the last product which an Alpha port of Red Hat 
Linux existed for.  Alpha architecture remained in the 
buildsystem with all rpm packages building for Red Hat Linux 7.3 
and 8.0 development also, however there is more to producing an 
architecture port than merely building rpms.  The majority of the 
work is kernel/gcc/glibc/toolchain and similar, and that requires 
serious engineering resources.

Any architecture that does not bring in revenue on it's own to 
cover the costs to produce the port in the first place is just 
not viable to do, because engineering resources are finite, not 
infinite.  Spending 30% of engineering resources to do work that 
brings in negative revenue is not something that increases 
shareholder value.

As such, ports to architectures that do not generate revenue, or 
do not have projected future growth potential to make efforts now 
reap profit in the future, are a waste of engineering resources 
that could be redirected to engineering features and enhancements 
on products that do actually produce revenue.

So if you see a particular architecture not being supported, in 
general, that means:

1) The cost to produce the port to that architecture is greater 
   than the amount of revenue that would be received in return 
   from doing so.

2) The engineering resources that would potentially be put into 
   developing a port of that architecture are better spent 
   working on features and other enhancements for the products 
   that do actually generate revenue.

3) No IHV or other 3rd party has contracted Red Hat to do a port 
   to that architecture.


For the case of Fedora Core however, it is a bit different.  
Fedora Core is developed primarily for x86, however the rpms 
continue to be built on all 7 of the architectures supported by 
Red Hat Enterprise Linux.  If a build fails on any one of the 7 
architectures when building for Fedora Core, then the entire 
build is failed and rejected.  A package is only accepted into 
Fedora-devel aka. "rawhide" once it has successfully built on all 
7 architectures (unless it is arch specific of course or has 
special circumstances).

So the majority of the rpms in the distribution are already 
architecture clean, both 32/64bit clean, and big/little endian 
clean, because we build on:

32bit LE: x86
32bit BE: ppc, s390
64bit LE: x86_64, ia64
64bit BE: ppc64, s390x

The major work that needs to be done for a port to a new
architecture, is:

- kernel, and all kernel features supported on other arches
- gcc
- gdb
- glibc
- other parts of the toolchain
- X11
- The installer (anaconda)

Fedora Core 1 is currently available for x86 and x86_64.  
Fedora Core 2 test 2 is currently available for x86 and x86_64, 
with some number of people interested also in ppc and sparc, 
however I'm not sure how far along these ports are, or wether 
they will become official FC2 ports or not.  Fedora is volunteer 
driven, so if someone wants to see a port, then they need to join 
the fedora lists and volunteer.

That does not preclude Red Hat employees either.  There are Red 
Hat people spending personal time working on ppc, sparc, and 
Alpha, however since Fedora Core for ppc/sparc/alpha are not 
going to generate revenue for Red Hat, it is not unreasonable for 
Red Hat to not spend paid for engineering resources on porting 
things to these architectures.  That does not of course stop 
various employees from doing things on their own time however, 
but it does make employee contributed bits for these ports 
limited to employee volunteered personal time.

In short though, if you want to see an architecture supported 
that is not something that is profitable for a company like Red 
Hat to spend money, time and manhours developing, we welcome 
volunteers working on the Fedora Project, and many of us will 
gladly volunteer to help out as well.

Back to your comment though...  I don't see what Red Hat Network 
has to do with this.  ;o)



-- 
Mike A. Harris       ftp://people.redhat.com/mharris
OS Systems Engineer - X.org X11 maintainer - Red Hat