Full LTSP-cluster or just load balancer for many thin clients

[ssanders at mbsbooks.com]

Hi all, first post to this list, a quick intro and a question. I love
Linux, erased my last personal FAT32/NTFS partitions 12 years ago and
haven't looked back. I have built several small LTSP deployments over the
last ten years for various charitable organizations, and is in daily use at
home. I have never had more than 10 clients. I found the K12OSN group, and
followed them from Redhat through Fedora and Centos. I have been using
Debian-esque OSs for the last four years, and have begun experimenting with
Edubuntu. Great work all around, and the 12.04 LTS is certainly appealing.

I have been a thorn in my workplace's side for some time, promoting the
concept of LTSP. I may have a chance now, and have potential hardware to
use.

All of my LTSP experience has been with rather basic setups: 2 nics, one
box, several drives, that usually has Squid and basic desktop apps for PXE
thin clients only (old PCs). For larger deployments I know that throwing
more iron at it is usually the thing to do. One server for apps, another
for /home, perhaps another for DHCP.

My possible installation at work would be a very light need for apps, but
about 180 thin clients. There will be no need for localapps, and no thick
clients. 90% of them only do two things: a character based 5250 emulation
(basically telnet, with some special function keys defined) to IBM
AS400/iSeries servers and a SSH connection to an AIX server. The other 10%
will additionally have Firefox to only look at intranet based reports.
There will never be any YouTube, Flash or multimedia use, and perhaps only
5 of them will ever have any local docs (and LibreOffice) at all. Only a
handful will ever need outside internet access.

XFCE instead of Unity or gnome-session-fallback would be fine, and further
reduce requirements. In testing just a few on Edubuntu 12.04, old 128meg o'
ram junkers perform just fine, some will be PCs removed from office use,
some will be little 15-watt fanless PCs.

Normally splitting up the services into different boxen would be the thing
to do for this many clients, hence my interest in ltsp-cluster. I will
perhaps only have two machines for the server however, and it needs to be
high availability. I don't have the specs on the servers yet, but they were
fairly high-performance two years ago. I can have RAID on them for disk
I/O, and gigabit (or several of them) to feed the clients' upstream
switches (most of the clients will be 100mbit).

Can I possibly run everything in one box, and have the second for backup or
fallover? We can live with normal HD deaths via hotswap, but this would be
a production environment, and need to pretty much function 24/7.

If the clients are set in BIOS to powerup after losing power, I know we
could have a boot-storm of network traffic if the power blips. The servers
will have good UPS, and the building has a big generator and enough batts
to hold it in the seconds it takes for the generator to spin up.

TIA-Scott S.