[NTLUG:Discuss] thin clients, are they practical and satisfactory?

[Bryan J. Smith]

On Tue, 2004-09-14 at 08:31, Will Senn wrote:
> All, Have any of y'all implemented a thin client setup that you are happy 
> with?

I worked with Florida Hospital on such a setup to replace their aging
"green screen" setups with a SCO OpenServer 3 back-end.  We used XDMCP,
IceWM as the WM and ROX-filer as the SM/FM.

I tried to get something going in my public school district when I as a
teacher, recycling 486 and Pentium systems as X-Terminals.

> I have been tinkering, without committing to the project of 
> setting up a server and client, but I haven't really gotten serious 
> because I am doubtful of the outcome.  If you wanna opine, here are some 
> questions to consider:
> What do you think is a reasonable thin client machine?

Anything that boot up to a basic X.  You can then connect via XDMCP so
_everything_ runs on the server.  The only kicker is encryption.  I've
yet to get SSH tunneling working with XDMCP

> What do you think is a reasonable server setup?

If I had to buy one for 100+ users:  

- HP Proliant DL585
  - (4) PCI-X Channels (Dual-AMD8131 chips)
  - (4) Opteron 800
  - (4) Dual-1GB DDR SDRAM (8GB aggregate, 4x128-bit/CPU-NUMA)
- (2) RAID-1 Card/Arrays (3Ware ATA or LSI SCSI)
  - Each card on its own PCI-X channel (on same Opteron/AMD8131)
  - LVM2 Stripped RAID-0
- (2) Dual-GbE NICs (1000BaseSX)
  - Most 1000BaseSX cards have a large SRAM (typically 0.5-2MB)
  - Minimal optical cost is worth the signal integrity in closet
  - Each card on its own PCI-X channel (on same Opteron/AMD8131)
  - 802.3ad Link Aggregation to Switch

Design:
- (1) Opteron handles all storage (up to 2GB memory mapped)
- (1) Opteron handles all network (up to 2GB memory mapped)
- (2) Opterons left, dedicated with 2GB each for processing
  (other Opterons back-fill when not doing I/O)

Should cost around $15-20K, depending on options (OS, GbE Switch,
etc...) -- not much more than a dual-Xeon MP (and cheaper than a
Quad-Xeon MP) solution.  HP is a core Intel IA-64 designer (partnered
with Intel), and even they are selling Opterons out of its sheer
performance and cost.

Xeon cannot match quad-Opteron's scalability -- especially at I/O.  Plus
it is a true, physical 48-bit platform, whereas even IA-32e "64-bit
extensions" Xeon is still 32/36-bit PAE outside the chip ("software
driven I/O for safety" -- until they merge the Itanium and Xeon
module/platform in a few years).  If you stay under 4GB, the hit is less
for Xeon, especially for only dual-processor, but Opteron still wins
well.

Choice whether or not to go with Red Hat Enterprise Linux x86-64 or
Fedora Core x86-64 should be left up to you.  Both will give you full
NUMA, I/O MMU and processor/IO affinity features -- especially kernel
2.6 (only Fedora Core 2 at this time, RHEL 4 will come about soon).

> What about just setting up linux server and using xdm/kdm/gdm to manage 
> user sessions from a workstation that will run apps on the server?

Running XDM, KDM or GDM directly on the workstation puts more load on
it.  It's up to you.

> Is that fundamentally the same (just a beefier thin client), or is it 
> radically different, or is it somewhere in between?

The whole term "thin client" is rather "new age."  X terminals varying
in approach, but most simply bring up X (from EEPROM, netboot, disk,
etc...), and then use XDMCP to run _everything_ on the remote system.

A "workstation" is one where a hard drive is required, and runs some
things locally.  Microsoft's use of the term, and proliferation now
among IT professionals, "workstation" is completely "fat" -- 0 remote.

> Just trying to get my mind (and my available hardware) wrapped around 
> the idea of thin client technologies as they relate to linux, after 
> reading Marcel Gagne's business linux migration book.

It was never much of a migration for me.  I was already supporting UNIX
and running non-Microsoft applications on Windows that had non-Windows
ports.  That was the mid-'90s.


-- 
Bryan J. Smith                                  b.j.smith at ieee.org 
------------------------------------------------------------------ 
"Communities don't have rights. Only individuals in the community
 have rights. ... That idea of community rights is firmly rooted
 in the 'Communist Manifesto.'" -- Michael Badnarik