BSD is dying

[Paul Richards]

On Thu, Jul 28, 2005 at 03:34:05PM +0100, Robert Watson wrote:
> 
> On Thu, 28 Jul 2005, Paul Richards wrote:
> 
> >There's a difference between being depressing and taking a realistic 
> >view of what the state of play is.
> 
> So maybe the real issue is that while there was a bubble in OS interest as 
> consumer and low-end hardware outpaced the OS vendors abilities to ship 
> software that could make use of it, there has now been a market correction 
> and the OS returns to being something that systems people care about, but 
> not so end users?

I don't think end users have ever cared. They used to run Windows, now
the more IT aware are seeing alternatives in OS X and Linux becuase it
allows them to do the things they always did, read mail etc and offers
them more power as they step towards having more control of their tool.

> I think the question you're asking is really "Can or should something be 
> done, or is it Too Late?", and specifically, are people interested in BSD 
> motivated to do what the Linux crowd has done -- swimming upstream and 

Essentially yes.

> not ever been convinced BSD ever did this well, and I always thought that 

No we haven't. Underlying my question is whether BSD ever can be a
mainstream OS or is it a niche hobby for the more technically
aware/competent OS hacker.

> the Linux people didn't have a leg to stand on, but given that they were 
> quite effective, I suppose a leg to stand on isn't required, just 
> enthusiasm.  Linux 2.6 is dramatically better than Linux 2.4, but the Hard 
> Linux Sell really too place based on the 2.2 and 2.4 kernels, which had 
> pretty terrible problems.

Whereas linux was saying "look we're a fully functional free unix,
don't worry about how you get it just make use of it", in contrast we
were saying "look, we're a fully functional free unix, you can build
it from source as one complete chunk, isn't that great". Most
non-developers really don't care how wonderful a development
environment it is. That includes sysadmins. That attitude hasn't
changed much in 10 years, if anything it's become more insular.

> I'm willing to give it a try, but I've been putting my energy into the 
> core parts of the OS rather than other parts because that's where the 
> foundational work needed to take place to make FreeBSD 6.x a desirable 
> place to be.  And I'm happy to tell people about it here, but since when I 
> joined there was little noise, I didn't feel it was the place to talk.

You turned up at a quiet time.

> systems.  FreeBSD has done very well in embedded, in ISP spaces, and so 
> on.  We've done less well in clustering, and other systems have been 

Have we actually done that well? I don't mean to be negative here
I'm trying to take stock. We used to be very strong in the ISP
market but I'm not so sure anymore. As for the embedded space, do
we really have a long list of products we can point to and say
they're powered by FreeBSD? Mac OS X and Yahoo are the only examples
in each category that I can refer to these days that people have
heard of.

I think the decline in ISP support comes down to poorly supporting 2
key technologies.

1) Poor mysql performance, I know you've looked into this in some
detail.  Personally I'd use postgresql myself but it's a simple
fact of life that if the herd goes in a direction and you don't go
with it then you get left behind. mysql didn't stack up on FreeBSD
compared to Linux and as ISPs started using mysql more they migrated
machines to Linux to run it on.

2) Java. A lot of big web projects are now J2EE based. The major
appserver vendors all support Linux as a primary platform so there's
no risk using it as the OS underneath your enterprise app. Not only do
vendors not support FreeBSD but until recently we didn't really
support runnning core java.

Java is a killer blow in the corporate market too, while we could
have made some headway as Solaris/AIX/HPUX lost favour in the big
iron machine rooms of corporate companies we just didn't support
the apps that they needed to run on top of the OS. We can't even list
Oracle and that's generally the starting point for corporate systems.

> - Taking PXE upgrades/installs from "possible but you have to know what
>   you're doing" to "trivial".

Not sure that's such a huge requirement but it wouldn't do any harm.

> - Significant work into remote management tools and services.  SNMP
>   integration is getting better, but is probably not what we need.  Since
>   I don't have much information in this area, I don't have a
>   recommendation.

Work on Linux, with things like Nagios has helped FreeBSD along here
but the momentum is again very much with Linux since Nagios grew from
the Linux user base and is largely developed there.

> - Work to allow bundling and re-bundling of update and version management
>   components, as well as work on automated upgrade paths.

This is critical for admins, they don't want to have to build the OS,
they just want to get boxes up and running and keep them running.
Difficult for us to do without commercial support though, and without
the expectation of long term commercial support sysadmins probably
won't risk adopting it.

> - Additional work to flesh out our directory service integration -- 5.x
>   brought in NSS, OpenPAM, Kerberos5 by default, etc, but while the
>   ingredients are there, the bundling and ease of use isn't.

I agree with this, we neither offer a coherent directory service nor
integrate well with others.

> - Integration of an advanced distributed file system.  If necessary,
>   development of the necessary pieces to make it happen.  I've always been
>   a big fan of AFS, but they've always had the same problem BSD has had:
>   superior technology bundled in a hard-to-use interface, developed,
>   managed, and promoted by people who are techies and kernel hackers at
>   heart.  But maybe the NFSv4 bandwagon is the place to be.

Mostly in the corporate world you see one of two setups, NAS and SAN.
I don't see much call for anything else. NFSv4 is obviously critical
for the former, we don't really have any SAN support at all. We don't
even have decent support for more basic things like volume management.

> - Some cleanup and retrofitting, replacement, or other activity of the
>   package management system.

It's starting to look a little shabby compared to many out there now
but we have a lot of ports committers so it's not a shortage of
resource that's the problem here. It's an acceptance that what we have
is ok because as developers we can make it work well enough.

> - The Management Tool.  No, not a web management interface, or SNMP. :-)

Well, that depends what that tool does :-) However, I would say I'd
expect it to be able to install device drivers for instance and that
requires more radical thinking in how we deal with loading drivers in
the kernel. Because ease of use from an interface perspective is not
what drives our kernel design I'm not sure we'll have see the will to
make that happen.

There's a more generic issue too. We don't ship a "system", we ship an
old fashined unix OS and make available lots of packages. If you
compare that to a Linux dist or OS X, then when the install completes
you have a functional system. By that I mean that they make decisions
about what they think their users want and get the whole system set
up. The GUI works, printing works, mail, web services all work etc.
That allows them to integrate these components much more because they
consider them part of the supplied OS.