[sf-lug] Large(ish) scale LINUX installations on older (x86) hardware
Michael.Paoli at cal.berkeley.edu
Sun Jun 24 14:55:16 PDT 2007
Well, many good points have already been made along this "thread".
I'll try not to say too much, for fear of increasing the fan-out ratio.
A few (not so?) "random" bits:
While clearing out some older e-mail, stumbled across:
Most notably the parts under "Installing Debian without CD-ROM",
including useful bits such as:
Various good points about "needs" assessment have been made.
I'd think key among them, is who will use them for what,
what are the objectives, musts, shoulds, cans, should nots, and cannots,
how will they be supported, and will they (hopefully) become mostly
self-supporting (rather than a project that whithers and dies due to
lack of support or unsupportability), etc.
I don't see it as being probable to be a simple marathon install - likely
a fair amount of planning - and even advance testing - may be required for
most optimal results. Even doing some testing - e.g. read-write testing
of disks - before attempting installations, might be useful - but that
takes time - and can be quite a bit of time with larger hard drives or
slower older hard drives.
I'd guestimate it may be highly useful to collaborate with folks that
have already done similar successful projects. What they've already
learned from that experience, testing, etc. would likely be highly useful
and generally actionable information. Some of the stuff Christian Einfeldt
has been doing with/in schools is one such project that comes to mind, and
is relatively local. There may be other projects that can also be learned
from (some of them might have lots of the relevant information on-line).
I'm not a huge fan of "cloning". Some things should not be cloned (e.g.
items which should be unique - such as host's ssh keys, filesystem
UUIDs, etc.). "Cloning" also generally only works quite well on highly
matched target hardware. Also, as disks get bigger and bigger, "cloning"
becomes generally increasingly time inefficient. E.g. writing out all the
blocks on a >80 GiB drive is rather time consuming; for a brand new
installation, even rather fully loaded, probably less than 10% of those blocks
actually need to get written to the hard drive; i.e. probably much faster to
plop a partition table in place, format the filesystems & swap, and then only
write out the needed filesystem data - e.g. via relatively "normal"
installation, or by writing files out from an archive (such as tar or cpio).
Such methods can be much faster than doing dd or the like - again,
particularly as drive sizes increase.
Also, with large numbers of systems,
what is often most important is not machine time to install, but
person time - set it going, ignore it for many hours, then come back
later, and see that all was done successfully. Whether it's 30 minutes per
box, or 72 hours typically isn't a major factor - as long as all ends well
and requires zero interaction after kick-off is set in motion.
disclaimer: I'm not necessarily pushing Debian as the optimal solution in
all cases or this specific case, ... I just "happen" to like it lots for
various reasons, and am typically more familiar with much of it than
similar aspects of other distributions.
More information about the sf-lug