Philip Prindeville wrote:
It depends. The disk serves to purpose. Restoring a clobbered machine
for an existing (and experienced) user. Or doing a first-time install
for a
new hire who can't find his posterior with both hands.
In the latter case, the more turn-key, the better.
Both can be set up off the network.
The network won't suffer from versions you wish you never released, once
you fix the problem.
I've not timed an install on current hardware, but I used to install
RHL 7.3 in under 15 minutes off a LAN.
The read rates on a DVD are comparable to network speeds (realistically).
I think you're more limited by processing and local disk writes... On a
Dell
L610, an install takes 20-25 minutes.
That sounds slow to me.
But I'm also installing a fair number of packages. Which reminds me of
As did I.
a couple of issues. (a) is there an easy way to figure out what package
group an individual package belongs to in an automated way, (b) does
the package name have to occur immediately after the group it
belongs to for inclusion/exclusion? (i.e.:)
@base-x
-sendmail-cf
I've never noticed any problem with ordering. However, latest Anaconda
has had radical surgery such that my experience isn't relevant.
and (c) how do you force a package to be omitted, even if something
else depends on it? For instance, NetworkManager requires
wpa_supplicant, but the wpa_supplicant on FC5 doesn't support
madwifi (the Atheros chipset that some of our laptops use)... so I
don't want to install it... (arguably, NetworkManager should be
able to install on a machine that doesn't have wireless PERIOD
without requiring wpa_supplicant... but that's another issue)... but
I do want to set up the ATrpms repository and pull down their
version of wpa_supplicant and install that instead.
I think that's supposed to work, if not with the current latest
Anaconda, then with a near future version.
On a slightly off-topic sidebar: I've noticed that yum will grab
the latest version of a package, regardless of the kernel you are
using... For instance, if I'm running kernel.2.6.16-1.2096_FC5,
it will still grab "kernel-devel.2.6.16-1.2211_FC5" if that's the
latest. Similarly for madwifi-kmdl packages, etc. Is that supposed
to be how it works? Seems broken.
I don't think the running kernel should be a factor. What if I install a
new kernel and some other stuff prior to booting the new kernel?
Anaconda isn't the only way to deploy Linux, there are also
third-party solutitions such as System Imager which is based on the
notion you install one system, get it "just so," and then clone it.
Unfortunately we use more than one type of computer... Ideally the
Shouldn't matter. Gernerally, all drivers get installed and
{re}configuration is done at (or after with hotpluggable stuff) boot time.
There may be some concerns with third-party (eg ATi an nvida graphics)
drivers, and where there are choices to be made between alternatives,
and disparate network devices where you have (eg) multiple NICs, but
mostly I expect any disk to work in pretty much any computer that will
run the kernel.
scripts will detect the computer type and customize themselves...
which makes me think that having lspci or dmidecode run and pass in
environment variables for the motherboard, etc. would be cool... and
avoid having the user have to parse that all out himself... And in
different groups, different people have different software installed
depending on their role.
You can have an array of golden images, oe a standard set of additions.
I can imagine different groups having different software requirements;
those could be handled in Anaconda by loading custom ks files from a
web server, and the web server could use CGI (or similar) to generate
the appropriate setup:
ks=http://ks.example.com/cgi/redfish.ks?department=accounts&essid=watsit&wep=s:bigsecret
or whatever
Or by looking up the MAC address => user name => user requirements/
user profile in LDAP as I mentioned previously.
Use of MAC address, seems to me, prone to error and to require more work
than is sensible.
If you have a "box" of peecees available from which you dispatch one to
the user, then the IP address it gets on booting _can_ identify where it
is (depends on network topology) and so inform the choice of software:
if it's not on Accounting's network it doesn't get Accounting's software
Remember, the Mac address identifies a network interface card, not a
computer, and a computer may have no NICs, and it may have several.
Further, and omboard NIC may be unused (some mobos have more than one,
or it may have failed and be replaced with a card).
btw If you examine the boot disk, you will probably find it's not very
difficult to start Anaconda yourself. I imagine you could have a
preliminary procedure that allows a users to identify themselves in the
usual way, maybe provide further info regarding special requirements,
use that info for them to "save to disk" a kickstart file.
For simplicy, a web interface featuring links (which does graphics on
framebuffer devices and which is _not_ shipped in FC) or elinks (which
only does text and is the browser you get when you start "links" in FC)
could be used in conjunction with a web app.
Note that wireless (and lots of other) configuration (and extra
packages) can be don in %post using tools such as sed, cp, mv and
grep. If you need to ask questions, look at dialog and xdialog (there
may be more variants too).
Couldn't find xdialog... I guess it's not part of the FC5 distro.
Anyone have any examples of using either? I suppose python+gtk
could also be used...
Debian uses dialog and its kin extensively. RH uses (or used to use)
NEWT but when I looked, documentation was scarce.
If you're building a customised distro, adding the rpms shouldn't be hard.
I suppose you could use GTK if you can guarantee a GUI environment.
btw I'd be reluctant to put user-specific information on a machine
(except a laptop): access to network facilities should require a
network (such as LDAP/AD) signon. On Windows, we have users' home
directories on a server, and they're cached on the PC the user logs in
on. If they use a different PC next time, that's fine.
Unfortunately, these laptops go "off campus" to homes, airports, customer
sites, etc.
Well, some user-specific information is going to be required to access the
corporate network if the user is off-campus at the time and wants to access
the Intranet via VPN, or SSL, etc.
You need to identify the laptop before allowing it to access the
coporate network, but never assume that identifying the laptop is the
same as identifying the user.
Which reminds me: we use Squid on-campus, and have proxy settings that
have to go into a dozen different places (wget, yum, Firefox, Thunderbird,
Opera, Evolution, etc). Why hasn't some bright spark come up with a
standard Linux/Freebsd libproxy.so that uses a single set of system-wide
settings and patch all of these applications to use it?
We use Squid, and we use transparent proxy. Our Linux and Mac machines
don't need any proxy settings.
For reasons I don't understand. IE doesn't always work with my settings,
but I can use AD to enforce proxy settings on Windows.
--
Cheers
John
-- spambait
1aaaaaaa@xxxxxxxxxxxxxxxxxxxxxxx Z1aaaaaaa@xxxxxxxxxxxxxxxxxxxxxxx
Tourist pics http://portgeographe.environmentaldisasters.cds.merseine.nu/
do not reply off-list
