Using Python for processing of large datasets (convincing managment)

[Alex Martelli]

Thomas Jensen wrote:
        ...
> However the CEO (small company, told you :-), made a couple of somewhat
> valid points against it.
> 1) He was worried about getting a replacement devlopper in case I left.
> 2) He said, "Name 3 companies using Python for key functions"
> 3) He was worried about the stability/reliability of python in our
> production environment (you know, 99.999 % and all that)

Come visit www.python-in-business.org -- we founded the Python
Business Forum, a non-profit alliance of firms which do use Python
"for key functions", and one of the PBF's primary purposes is to
help reassure CEO's (and CTO's &c, for larger companies:-) about
just such issues as these.

Point 1 is basically never a problem because Python is to easy
to pick up (and somebody else's Python code is easier to maintain
than for any other language), but the PBF aims to help, in the
long run, by establishing a network and referral point for
Python trainers and consultants.

Point 2 is easy -- Zope Corporation relies on Python so much it
pays to employ the Python core team, Google uses Python enough to
demand Python skill in some of its job offers, Fidelity (a huge
financial trust in Britain) is dependent on Python (tbe basis of
all of ReportLab's excellent products) for all of its reports-as-
PDF web sales strategy -- that's three.  We can no doubt find
more, but three is what your boss asked for.

Point 3 is troublesome if taken literally -- I know of NO language
claiming 99.999% freedom from defects for its implementations.
The standard Python distribution is certainly far less buggy than,
e.g., most Microsoft language products I've ever used -- on which
many corporations, wisely or otherwise, rely for their business --
but that's still a FAR sight from 1-part-in-100,000-or-lower
defect rates.  One of PBF's plans, named "Python in a Tie", is
a very stable Python distribution, due to be blessed by Guido but
tested very intensely by the PBF.  But the targets, though not
quantified so far, are NOT as ambitious as 99.999%.  If you're
planning deployment in life-critical applications (the only
reason I can see for such strict, enormously expensive demands)
I think, regretfully, that you'll have to look elsewhere (I'll
be curious to hear about what software tools you find that claim
those reliability levels, at least if they back it up with real
hard-cash reliability insurance -- words are cheap).


Alex