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

Re: Proposal for new architecture support/distribution


On Mon, 1 Feb 1999, Daniel Jacobowitz wrote:

> But assembler for one powerpc should work on another.  If it doesn't,
> then it should be fixed.  We have a working mpg123.

Indeed, but it won't work on an RS64 II, or a Power2.
> Processor is not the issue.  That should be public information.  They
> ARE the same architecture.

Incorrect assumption. The PowerPC RS64 II is not the same as a Power2 is
not the same as a PowerPC With X5 Cache is not the same as a PowerPC
(603,603e,604,604e). Comparing the PowerPC RS64II purely on the
architectural level with a PowerPC 604e, is akin to comparing an
UltraSPARC II to a PentiumII.
> I'm confused.  Different kernels should _never_ harm userland
> compatibility.  Only the kernel should ever need to know the difference
> between UP and SMP.

It's not the userland/kernel that I'm most worried about; more the
userland applications themselves. If a multi-threaded application is
compiled UP, then in most cases, it will not take advantage of an SMP
system, at least in my experience.
> Plus, why do you claim that the server market is not the Linux focus? 
> We go much deeper into the server market than the workstation one.

Perhaps I should have been a bit more specific. We all know that Linux is,
to put it bluntly, blowing NT out of the water left and right. However, I
personally don't see much headway at all being made into the workgroup
server and database server markets. The world knows for cold hard fact
that Apache is the most popular and most widely used web server on earth.
However, I personally don't know of any *real* statistics proving what is
the most popular database on earth. (SQL is not an answer, folks. SQL is
the language; not the engine.) Furthermore, we have Samba, we have NFS, we
have Coda, we even have NCP. 

I honestly cannot say I have seen a Linux system acting as a fileserver,
or a workgroup server of any sort, in the sense of handling user logins
and home directories, as well as applications. Proxy servers don't count.
(I stopped counting Linux proxy servers *long* ago. They're everywhere.)

Now, moving on to what Jim brought up.

On Mon, 1 Feb 1999, Jim Pick wrote: 

> You'd have a separate RS/6000 kernel which would be compiled for SMP.
> This should have no impact on the existing PPC userspace.
> The userspace stuff shouldn't care if it's running on UP or
> SMP. Remember, i386 supports UP and SMP with one userspace.

This is defintely a valid point. As I stated, I'm more concerned about
userspace applications. Furthermore, let us not forget that there are
other things that simply *must* be done to the kernel to make it run
acceptably on RS/6000 systems that are PPC (603,603e,604,604e) based.

> Debian already has all these applications (except for Oracle).

*grin* I don't think getting Oracle in on things could be too incredibly
difficult, with some finegaling and negotiating.

> Basically, all Debian distributions look the same, regardless of the
> underlying architecture - because they are built from the same set of
> sources.

This is a very strong point. One that I had to think long and hard on. And
it led me down a new train of thought.

First off; I'm going to have to stick by a seperate RS/6000 distribution
due to the kernel work that must be done, and possibly other userland work
that I honestly don't believe any one of us can anticipate anywhere near
accurately alone. Furthermore, the name alone will likely bring along
further supporters, especially IBM, who will likely make donations to
Debian, in the form of hardware and funding. We all know that this is what
keeps things going, therefore, it cannot be a Bad Thing(tm). ;)

Therefore, I propose that we begin work on Debian for the RS/6000. The
distribution will be no different from any other distribution, except for
the architectural differences. Depending on how things are done, we may
have to subdivide it by architecture, ie; Debian-PPC, Debian-PPC64,
Debian-PPCX5, Debian-POWER2. Personally, that's about the *last* thing I'd
like to see. Much more work, and possibly more confusion.

So, I propose that we start simple. UP to 8 way SMP RS/6000's, based on
the PowerPC (603,603e,604,604e), with PCI/ISA only. While this may sound
like it's cutting out many RS/6000's, it really cuts out very few. What I
would like to see is a stable Debian-RS/6000 for machines meeting the
above requirements first, before we delve into the unknown. In this day
and age, it's easier to get backing when you already have it working.

Secondly; I propose the beginnings of a Debian offshoot. Now, something
tells me that this will be met with a great deal of resistance (from the
people who brought you vrms.deb;), but I believe it may be in all of our
best interests, to further garner market approval and acceptance of Linux. 

Debian for Business. The only difference is in the packages, and install
methods. I am totally opposed to the route that RedHat has followed,
charging ~$40USD for their distribution. I do not propose that we do that.
Not in a million years.

What I am proposing is that we create an offshoot of Debian, that is
designed specifically for businesses, both large and small. With all the
packages of the current Debian, perhaps additional packages that are
business focused, and an install system that the traditional 'harried IS
guy' won't have anything complain about. 

This is a dream I've had for a while myself, that would fit in perfectly
here. Some businesses are expressing the desire to move to Linux on the
desktop. And move all their servers to Linux. Sometimes, there's more than
just a few desktops. And sitting through an install of every single one is
just not something you want to do. You want to have all your specific
packages pre-selected, so that you just pop in the bootdisk, hit enter,
and walk away as it installs over the network. 

Now, I know that I'm already going to get the 'we already have something
like that' messages. And here's where I'm going to explain where that
argument is wrong. 

We have 'generalized' installation profiles. What I propose is something
totally different. What I am proposing is a bootdisk, that contains a list
of packages to be installed, that is created by the 'harried IS manager'
using either dselect, or another tool we could always come up with. 

I do not propose the abandoning of 'generalized' installation profiles for
Debian for Business either; simply the changing of the generalized
installation profiles to more business oriented ones, such as 'Webserver,'
'Samba Server,' 'Database Server,' 'Application Server,' etcetera. 

Granted, someone who knows Linux, should be able to do this on their own,
but not all companies switching to Linux really know that much about it.
Nor do they have time to fiddle with getting it installed. Doing this
could give Debian a serious edge over RedHat in the business market. 
(Which I personally do not like. ~$40USD for 'free' software? No thanks, 
RedHat. I'll stick to something that really is free.)

As always, feedback, input, etcetera are always welcome. The whole purpose
of mailing lists and such is so everyone can bounce ideas off of everyone
else, afterall. ;)

- -Phillip R. Jaenke (prj@nls.net | InterNIC: PRJ5)
 "Look. It works this way." "Why?" "Because the designer said so."
 "Why?" "Because the designer is a moron. Let's fix it." --anon.

Version: 2.6.3a
Charset: noconv


Reply to: