Is Yahoo! moving from FreeBSD?
tedm at toybox.placo.com
Fri Feb 25 09:11:59 GMT 2005
> -----Original Message-----
> From: owner-freebsd-questions at freebsd.org
> [mailto:owner-freebsd-questions at freebsd.org]On Behalf Of Daniel
> Sent: Thursday, February 24, 2005 8:35 AM
> To: freebsd-questions at freebsd.org
> Subject: Fwd: Is Yahoo! moving from FreeBSD?
> sorry, i should have sent this to entire list...
> On Thu, 24 Feb 2005 01:43:32 -0800, Ted Mittelstaedt wrote
> among others...
> > FreeBSD does not have some of the things - such as
> distributed management
> > of hundreds to thousands of FreeBSD servers over a large enterprise -
> > that
> > are a requirement for big companies.
> would not these things be worthy of implementing in FreeBSD?
It isn't a question if they would be worthy. There's worth in
anything in FreeBSD of some level.
It is much more a question that these sorts of tools are -very- complex
if they are any good, complex to build, complex to maintain, and complex
to operate. We are talking a tool that might take a few months of
by someone already experienced in FreeBSD to become proficient with. Or
tool that might take a year for someone not already familiar with FreeBSD
to become proficient with.
Furthermore your only talking a very limited market for them.
The model for this kind of tool is one where you have a handful of really
experienced developers who are constantly working on them, and selling
into a market of perhaps a couple hundred experienced admins in the
world, if even that. Between them these tools control thousands of
servers and desktops. That means, unfortunately, you have to extract a
fairly hefty amount of money every year from that group of couple hundred
experienced admins. You do that by licensing on a per-server basis,
basis, etc. Since the big companies can well afford this, it works out
fine, but only as a commercial software offering.
You cannot build these kinds of tools as a one-shot thing, or build to
solve a specific problem, and have them last.
> making a good OS that runs on cheap, low-end machines is nice, but the
> real money come from companies...
As has been said countless times in the past, the ideal Free Software
is one where you have a commons of core operating systems and general
purpose applications that are open source, and companies then contract
with developers to customize those applications to their specific needs.
The commercial software approach has always been for the commercial
to come out with a product that tries to do everything for everybody,
and as a result does not do any one thing that well, and companies
then modify their business processes to fit the software.
Both approaches cost roughly the same money - with the Free Software
you spend it in labor, with the commercial software model you spend it in
But with the Free Software model, you end up with customers getting
what they need. With the commercial model you end up with customers all
working the same way their competitors are.
> another idea, a study of what features big companies want from an OS
> should be conducted...by you, maybe or some other people interested
> and these features be prioritized for FreeBSD...
On the surface that seems like a reasonable way to get FreeBSD's usage
increased. But there are some major reasons this wouldn't work.
First, such a survey assumes that big companies know what features they
want. The reality is often a big company will see a new feature they
have never heard of before, never knew could even be implemented, and
once they now know about it, they want it. In other words, your better
with a small team of people who are gurus, have a huge amount of
in these environments, getting together and brainstorming.
Second, this approach assumes that if you presented a big company with a
OS that had every exact thing they wanted, that they would indeed switch
to it. In reality they may still not switch, for example they may not
believe your OS could do it, or the implementation problems would be
too difficult. Kind of like dangling a lollypop in front of a kid who
is on the other side of a 4 inch thick piece of glass - he would love
to have it, he would be jealous that it's there and he can't get to it,
but he still isn't going to be getting that lollypop.
Third, this takes the "does everything for everybody and not any one
thing well" approach. For example, you get 3 respondents, one wants
item a, item b, item c, one wants item a, item c, item d, one wants item
a and item e. You prioritize this and produce item a first, then item
c. But after all that labor still nobody wants it - the first respondent
can't use it because it's lacking item b, the second can't use it
because it's lacking item d, the third can't use it because it's lacking
item e. Another way of saying this is that while people can be
statistically profiled, nobody ever exactly matches the profile, and
thus in a situation where only an exact profile match will do, your
statistical analysis of the profile is still a waste of time.
Last, even if you could get all the features wanted by big companies
into FreeBSD, unless those features are implemented exactly the way
those big companies want them, there's still no benefit to switching
to FreeBSD. The big companies already have to compromise what they
want to fit into the Microsoft holes, for FreeBSD to be more attractive,
it must be a "better windows than windows" to use a slogan.
More information about the freebsd-questions