Linux kernel compatability

Jeff Roberson jroberson at jroberson.net
Thu Jan 6 22:04:59 UTC 2011


On Thu, 6 Jan 2011, Marcel Moolenaar wrote:

>
> On Jan 6, 2011, at 9:59 AM, Warner Losh wrote:
>
>> On 01/05/2011 21:00, Scott Long wrote:
>>> I'm sorry, this simply hasn't been true in my experience.  I've worked with companies that have decided to support FreeBSD, and I've worked with companies that have decided not to support FreeBSD.  Emulation has never been used as an excuse to not support FreeBSD.  It's purely a cost/benefit decision.
>>
>> Yes.  I've been on the inside of a few of them, even seeing some business case figures.  These usually say that for the segment that company X is going after for product Y can sell 1000 units to customer W and another Z000 to the market as it emerges over the next 2 years.   1000 units gets them $200k profit, development costs are $100k for developer time, test time, etc.  Z is large, so potential revenue form this project is in the millions, with a guaranteed small initial profit.  Decision: go.
>
> But one cannot ignore the fact that a compatibility layer allows companies
> to support FreeBSD at lower development cost by eliminating the native port
> and instead just focus on the qualifying their Linux support within the
> emulation layer. If decisions are purely cost/benefit, then a compatibility
> layer reduces the cost, hence increases the benefit so if FreeBSD is at all
> a consideration, it will be through emulation.
>
> Is this what we want promote?
>
> Also, the experience that you and Scott have may be biased. You won't want
> to work for a company that is inherently Linux centric, right? Likewise,
> Linux-centric companies may be more interested in hiring Linux hackers and
> not FreeBSD hackers, right? So, doesn't that mean that your experience is
> ipso facto biased towards the companies that would even consider FreeBSD
> to begin with?
> What about those companies that couldn't care less about FreeBSD? Those
> for which cost and benefit are absolutes?
> I very much doubt that they are going to invest in an entirely new OS --
> in order to support it natively, when their Linux-centric development teams
> can do the same using emulation?
>
> What I'm saying is this: do we really have an abjective view or are we
> biased towards FreeBSD-friendliness simply because we are FreeBSD hackers
> discussing on a FreeBSD email list and working for companies that like
> FreeBSD in some form or shape?

Hey Marcel,

You may know that as a contractor I work for quite a number of different 
companies and not always on FreeBSD.  I have worked with companies 
developing products from the start when operating system selection had to 
be made.  I have worked for companies with existing Linux products that 
want to explore costs to produce FreeBSD versions or drivers.  I have been 
at companies who bailed on BSD and switched to Linux at great expense.  I 
have bugfixes in the Linux kernel and have written Linux device drivers 
and other features.

I'm not going to claim that I have an objective viewpoint or that I know 
the right answers.  I have seen a lot though by virtue of working with 
many companies and evangelizing FreeBSD, often unsuccessfully. 
Additionally, I have good friends who work regularly on Linux, contribute 
to lkml, and are generally connected in that world.

My view is that companies needing a platform will reject FreeBSD for lack 
of features or lack of marketing clout.  Linux has a stronger brand and 
people may want to advertise that association.  IB is one feature that has 
cost us users, as is virtualization, and perhaps some driver support.

The desktop apps that people want are a non-starter.  If you talk to Linux 
people they're barely holding on to native versions of many packages with 
significantly more users than we have.  They complain that adobe isn't 
supporting them, flash is too buggy, they're second class citizens in 
every way in commercial desktop software.  To think that these companies 
even consider us is overstating our case.

We need to do almost everything we can to reduce the barriers of use for 
FreeBSD in server and platform (juniper, isilon, type users) users.  In my 
view, this is the core market and this is where we shine.  They only way 
to keep users, developers, and mindshare is to grow our base by providing 
what they need.  Emulation is one, imperfect, way to get there.  The ports 
tree is obviously exceptional and superior where possible.

In this specific case, there is no reasonable way to get an infiniband 
stack without the wrapper.  There is already an industry consortium with 
developers from many companies working full time to make the one in linux 
a reality.  There is only a casual interest in BSD and only companies 
already heavily invested in BSD are willing to go through the expense and 
time required to get a stack.  Others simply install linux and move on.  I 
can tell you that this was the only stumbling block to using FreeBSD for 
a company started by one of Isilon's original founders.

I believe FreeBSD is an organization of principals and values that make it 
excellent.  I understand how this is a compromise and even I myself will 
tell you it is.  However, we need to keep in sight that we won't all get 
to continue doing what we love if we don't keep working to make it as 
appealing as possible.  We are in every way the underdog the underdog 
doesn't have the clout to dictate terms.

There has been quite a lot of good discussion and interesting points of 
view presented.  However this will likely be my last contribution to the 
discussion.  I'm going to work with the appropriate people to get this 
reviewed and in and will do what I can to minimize any view of a generic 
linux wrapper to prevent the abuse in other new code.  I hope this 
resolution is acceptable enough to the detractors.

Thanks to everyone who participated for their thoughts,
Jeff

>
> Are we therefore the right people to argue whether Linux KPI emulation is
> good or bad for FreeBSD in the long run?
>
> I'm indecisive. It may be a damned if you do and damned if you don't
> kind of scenario. If that's the case, I'd rather be damned without it :-)
>
> -- 
> Marcel Moolenaar
> xcllnt at mac.com
>
>
>
> _______________________________________________
> freebsd-arch at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-arch
> To unsubscribe, send any mail to "freebsd-arch-unsubscribe at freebsd.org"
>


More information about the freebsd-arch mailing list