RE: Use of CVS on large scales

[Teala Spitzbarth]

Lucas,

I know this topic is fairly well beaten to death, but I thought I would
add a few words from my experience with both CVS and ClearCase.  I
worked at SCO as a build, integration and SCM engineer for almost 10
years.  When SCO purchased UnixWare from Novelle, they 'inherited' (for
a large sum of money) ClearCase (ehem, being freshly ported to
UnixWare), and I was part of the team that rolled ClearCase out to 200+
engineers that had previously been using SCCS or STAR. I don't even want
to tell you how long that rollout took to plan and execute, nor how much
the equipment cost to run CVS effectively on the source for the full
UnixWare product line (over 10 million lines of code). I'm currently
working at a small, 30+, startup company that is using CVS with no
serious hitches.

- Scott Pyatt's email was truly sound.  Any consultant that is not
walking you through a grounded evaluation process of several tools and
prioritizing your needs, imho is not earning their money.

- you don't say exactly how large your code base is or how many
engineers you have right now. You may need a different tool "later"
(scalability issues, features etc) - but what do you need right now?
How much money does your company have to spend on tools, hardware and
administrative costs now?  The approach I have taken in our company is
that CVS is the best tool while we are getting started.  When I last
worked on ClearCase it provided basic conversion tools from CVS. So when
the company grows to the point it truly needs a more sophisticated tool,
and has the budget to cover it, then conversion is most likely going to
be possible from CVS.  

- I want to echo the statements that have been made about importance of
networking.  The only problems I have seen with 'CVS' here, are really
issues of networking overload on our server and networking
infrastructure. Those issues are going to be orders of magnitude more
significant with ClearCase, at least in my experience at SCO. I've seen
basic 'full pull' processes under ClearCase balloon to taking 18 hours
or longer (I forget the exact implementation details that caused this).
Serious work had to done to write scripts and modifications to bring
these times down. Branch merges took hours, and could often have errors
if view configs weren't designed correctly. Developers definitely
grumbled about response times.  VOB/View servers were overloaded if more
than one team had to do a nightly pull in the same time span.  I doubt
anyone that makes a blanket statement that ClearCase is more 'scalable'
has ever worked with it on a truly large scale implementation.

- The administrative cost of ClearCase is significant.  How many
engineers are in your build team?  It takes time, (months), to learn the
tool, design your company policies about views, labels, triggers, get
the hardware in place and roll this tool out (especially if we are
talking about Multi-site). At SCO, for a very large codebase, it took on
the order of 1 1/2 years, with about 3 full time people and the support
of about 10 other build & integration engineers. Not trivial. And the
most advantageous feature, using clearmake, was not able to be
implemented given the large scale changes it would have required in the
source tree.  If you don't devise policies and views well in ClearCase -
then as John Minnihan pointed out you are going to loose days & weeks of
developer and project time. Are the administrative & developer training
costs really balanced against the features you are gaining from
ClearCase?

- Weigh in the balance when evaluating ClearCase that you want to run it
on a well established platform (i.e. not a recently ported platform).
John Minnihan mentioned the MVFS implications of ClearCase; fresh ports
can show serious system issues.  With the port that was done to
UnixWare, our team spent significant time escalating kernel panics to
our Escalations team and Rational's. Of course some issues were UnixWare
issues and many were Rational's, but from an SCM engineer's perspective
it is just downtime.  Umm, I've never seen CVS panic the kernel :-)

- One aspect of CVS that I have found weak and miss from ClearCase is
the ability to write effective triggers.  As far as I have been able to
figure out, especially when using pserver and WinCVS, the commit and
checkout programs are fairly limited in what they can do.

- I haven't used Perforce yet, but I know several SCM engineers that
feel it is one of the best tools out there considering cost and very
good platform availability for both server and client pieces.  If you
need more than CVS offers right now, I would be sure to evaluate
Perforce along with ClearCase.

Good Luck,
Teala Spitzbarth