gnu-arch-users
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Gnu-arch-users] Archives use cases


From: Clark McGrew
Subject: [Gnu-arch-users] Archives use cases
Date: Mon, 06 Oct 2003 13:32:37 -0400

Hello,

I've been looking at how to migrate a physics experiment from CVS to
arch, and I've run into some quandaries about the best way to structure
an archive.  For the moment, my puzzlement is centered on two use
cases.  I should mention that I'm thinking about this based on my CVS
experience and am surely missing the obvious answers.  But I suppose
that other CVS users are likely to stumble across the same sort of
problems.

Both of the cases are questions about how to manage the flat arch name
space when there are a large number of categories.  Does anybody have
experience with handling large projects?  I am aware of configuration
packages, and I see how they could help.  But, I'm not sure that they
solve the overall problem which is how to handle the large number of
categories.

The first case is how to use arch for a slightly more complicated setup
than a single library project.  I'm a physicist and write lots of talks,
memo's and papers.  I have been using CVS as a way to both backup the
documents I write on my portable, and to keep a history of everything I
write.  I do this by keeping a CVS tree

    doc
    doc/superk
    doc/superk/talks
    doc/superk/talks/doe_sep03/
    doc/superk/talks/doe_aug02/
    doc/superk/talks/noon_dec00/
    doc/superk/papers
    doc/superk/papers/arnps01/

This structure repeats for each experiment that I work on.  It works
pretty well since CVS lets me check out only the documents I need at any
particular moment (a requirement, I write a lot) and the tree structure
keeps things organized.  My initial idea is is to replicate the
structure with categories and branchs. 

   superk-talk-funding--doe--03.09
   superk-talk-funding--doe--02.08
   superk-talk-atmpd--noon--00.12
   superk-talk-tau--noon--00.12
   superk-paper-review--arnps--2001

>From a librarian point of view, this seems like it would work, but the
flat archive namespace seems a little problematic.

The second case is how to use arch in the context of a relatively large
physics collaboration.  A large collaboration might have several hundred
active physicists spread around the world, and about 100 people actively
developing software.  My guess is that this means a physics
collaboration would be considered a moderately small software
development project, but one that mirrors free software development
rather than commercial.  Physics software is usually divided into
several loosely connected libraries (packages), each package is
developed by a small core and has an "independent" released
schedule[1].  Putting this in context, an experiment like Atlas based at
CERN has something like 600 independent package.  

Once again, arch's flat name space seems to pose a challenge.  How
should I handle several hundred categories in a single archive.  I
suppose one possible solution is to break the code into multiple
archives, but "everything in one place" is working pretty well, and I'd
like to keep with that paradigm.  One extra problem is that the software
is generally "self-organizing" and we don't have "managers".  Most
collaborations don't have a single gatekeeper (like Tom, or Linus) to
handle over all structural issues.  The most organized experiments have
a software committee, unfortunately the edicts are often ignored.  That
seems to be a fact of life, and one that is not likely to change.

Thanks,
Clark

[1] We've have gotten pretty good at handling the dependencies between
the packages.  If anybody is interested, take a look at CMT.  It's a
build system that has been developed specifically for the physics
software environment.  I wouldn't claim that other fields should emulate
our development habits, but by going with the flow C. Arnault has
developed a nice build system that fits our idiosyncratic requirements.

-- 
Clark McGrew                    Univ. at Stony Brook, Physics and Astronomy
<address@hidden>        631-632-8299





reply via email to

[Prev in Thread] Current Thread [Next in Thread]