[LONG] Maintaining several identical installations of (B)LFS

Alexander E. Patrakov semzx at newmail.ru
Thu Jul 31 21:22:53 PDT 2003


On Friday 01 August 2003 02:52, Dagmar d'Surreal wrote:

> Okay, then perhaps you could have been clearer that you were building
> transplantable binary packages instead of building things from scratch
> on each machine.

Further clarification: up to June 15, we used NFS-root. From June 16 to now, 
the LFS installation is used by nobody. We are thinking of a non-NFS-root 
installation. It is clear that sharing /usr and /opt via NFS while having a 
local / partition will not improve things as compared to NFS-root since 
OpenOffice (in /opt) is the most important traffic producer (swap files are 
on NFS, but they are never touched during any normal work).

Some packages were compiled by me at home (since this NFS-root installation 
was basically a copy of my hard disk contents minus /home/* minus /opt/wine 
minus /opt/fakewin32) and transferred on CD-RW to the classroom as Slackware 
packages. I installed them on one machine and the changes propagated to all 
computers via common NFS-root.

Some packages (small ones and Emacs) were compiled there, and automatically 
got propagated to all workstations since they use common NFS-root. The 
resulting Slackware packages were sometimes forgotten there instead of 
writing them on a CD-RW and bringing home.

> Oh, on the point of hardware differences, if these differences involve
> PCI cards, installing hotplugd makes all that just "go away".

Yes, this is the solution. My solution involved probeall in /etc/modules.conf

> > > Umm... side note.  Most DHCP server implementations can identify
> > > machines by MAC address and by client name.  It would have probably
> > > been easier than cooking up some script.
> >
> > Yes, and my script just uses this fact.
>
> You do not need a _script_ for this.  Use ISC's dhcpd and man
> dhcpd.conf, paying special attention to the "host", "hardware", and
> "group" directives.
>
I don't need a script on the server. The script runs on the client before any 
mentioned directories are used and basically just does

<figure out $IP>
mount --bind /etc/local/PER_HOST/$IP /etc/local
mount --bind /tmp/PER_HOST/$IP /tmp
mount --bind /var/PER_HOST/$IP /var

and similar things (I put all machine-specific configuration files into 
/etc/local and make symlinks as needed). Of course, really it is more 
complicated - it creates these directories if they don't exist (or if the 
template was updated) and populates them with the contents of template 
directories.

> > > You won't buy a switch but in a later email you're glibly talking about
> > > gigE??
> > > Buy a switch.  Switches are good, m'kay?
> >
> > I said "probably", and I didn't mean that gigE will be used here. This is
> > an unchecked recommendation for others. BTW, with 7-8 workstations, I
> > would not ask questions and would continue to use NFS-root. But there are
> > 15 workstations.
>
> You really, really, really need to pester them to replace that hub with
> a switch.  That will clear up most of your bandwidth issues almost
> immediately.  (I'm a bit astonished you managed to find a 100bT hub with
> that many ports.  They're rare now because they reek.)

In fact there are two cascaded hubs. BTW if the switch clears up enough 
bandwidth, I will revert to NFS-root.

>
> > Also, I don't have right to buy anything and all recommendations from
> > admins get ignored here. They wasted money and bought that P4 monster
> > (now practically unused, only mail server with 5 users and a web server
> > with 10 visits a day) instead of a switch.
>
> If the admininstrator recommendations are really and truly being
> ignored, get you resume in order.  Seriously, that is unbelieveably
> broken.

I already tried posting my resume looking for additional jobs. Most important 
problems with it are the following:
1) Nobody needs UNIX or Linux in the city where there is a Microsoft filial 
agency. Linuxoids are called paranoid (since they trust nobody) and mad and 
people who trust nobody are directed to the psychiatrical clinic.
2) Secretaries typically know nothing beyond MS Word, so my resume in PDF gets 
the following attitude: He could not even make it readable, so why talk with 
this man who knows nothing.
3) It is practically impossible to find a job where the knowledge of a stupid 
Windows program "1C Bookkeeping" is not required, and others say that working 
around authors' mistakes takes the most of their time.

> To get _really_ kinky, you could start using BOOTP & TFTP and get your
> particular rootfs for Linux off the network, the location of which would
> be dictated by the particular boot image... although this is
> considerably more obnoxious to do with x86 machines than it is with say,
> a bunch of Sparc 5's.

I already use BOOTP, and Etherboot does not require TFTP at all - it can 
download kernels from NFS.

> Like someone else here said, a small local filesystem would be a really,
> really good idea.  You could probably get away with putting /, /tmp,
> /var, and swap on a local disk in under 384Mb of space using four
> logical partitions.  This would eliminate a number of problems for you,
> and mounting /usr and /home over NFS would then be a cakewalk.

Taken into account. But see the beginning of the message.
-- 
Alexander E. Patrakov




More information about the blfs-support mailing list