Teaming Thin Clients?

[AntiEarnie]

Not sure if this is a good spot for this one or not. We are planning using our 2003 servers as terminal servers. Fun thing is that its critical to keep the Thin Clients up and I am not having much luck finding one that can support teaming/Switch Fault Tolerance. Has anyone done something along these lines?

 

[terrywilson]

I have seen HP clients that support PCi cards that you could put a second nic into, but the trick would be in getting the OS to automatically fail over to it.

I suppose the next step up would be to build your own custom thin client, dual nics, probably with a linux distribution and then run packages like balance or something to manage traffic between the nics which themselves come back to different switches.

Be interesting to know your solution to this one. Good Luck.

 

[AntiEarnie]

I had already checked with HP on them (t5700 series). Problem is the bay does not have enough room to handle the PCI/PCI-X dual NIC cards that can team (onboard NIC does not support it). Has to be a standard PCI in size and I have yet to find one with dual NIC. Someone has brought to my attention an onboard dual NIC thin client but darn if I can find one of those that can handle teaming either.

By and large no Off The Shelf stuff is frowned upon.

 

[compuveg]

It sounds to me like you need to build some barebones XP Pro systems to use as thin clients, and team those NICs.

How can a thin client be that critical anyway? If it crashes, doesn't the session just hang disconnected until they reconnect?

 

[AntiEarnie]

compuveg-
We are using PC's now but have had several hard drives and power supply's go out us. Going solid state should get rid of the more common problems we have had for the last few years. The environment will be way more forgiving to them at the least.

If the application crashes its quick enough to recover but I can not recall having terminal server crash on me yet outside of communication loss, thus the hunt for teaming. As for critical, a computer is generally no more critical then the application it is running and these will be mostly runing industrial control/monitoring software. Managers and operators get, fussy, when they can not control the plant.

 

[terrywilson]

I wonder if a more cost effective / simpler solution would be to use Citrix session reliability / smooth roaming with cheaper thin clients (without dual nic functionality) that allow workers to move their sessions to different terminals if you encounter problems?

 

[compuveg]

It is also possible on Terminal Server to allow your clients to reconnect from a workstation other than one that they originally logged in from.

In the GPO, you can find the setting in the machine part, under Administrative Templates\Windows Components\Terminal Services\Sessions

The setting is

Allow reconnection from original client only

This corresponds with the registry entry

HKLM\SOFTWARE\Policies\Microsoft\Windows NT\Terminal Services\fReconnectSame

 

[jkupski]

I have to agree with compuveg when he points at cheap PCs running winXP.

That said, if your issue is HD and power supply failures, moving to thin clients will only solve half of your problems--you'll still be just as likely to have a power supply die. If the terminals are as missions critical as you're making them out to be, throwing hardware at it seems to be the best idea, IMHO--XP boxes as above, but with redundant power supplies and hardware RAID, in addition to the fault tolerant NICs.

BTW, as an interesting side note, once upon a time I worked on the dev team for a package similar to the one you're using (Siemens PCS7 OSx.)

 

[compuveg]

Well, Re-reading this thread made me think.

Ya know, you're talking about some serious control rooms it sounds like. Control rooms that control very critical processes. Depending on what these control rooms are, we could be talking about malfunctions causing anything from buildings to get so hot people have to work in their underwear to nuclear reactors melting down.

Something tells me that you aren't at either extreme, but regardless, you should remember that many versions of Windows software state in the EULA explicitly that the software is not to be used in applications where malfunctions can cause loss of life or limb, such as life support systems, etc. They don't state it exactly that way, but I would make REALLY REALLY sure that you're not going to kill anyone with a blue screen, or a black screen if the power supply croaks.

 

[AntiEarnie]

From all the digging and research it looks like it can be done if you are running with embedded Linux. You just need to add in the bonding driver and your good to go. However, in order to run the terminal server software we are planning the thin clients have to run a proprietary OS that cannot do this.


terrywilson -
Actually we are planning to use ACS ThinManager to get terminal server failover and a couple other handy features. This is more or less what killed the redundant NIC thing as the thin client OS doesn’t support teaming/bonding. Dual subnets, yes, but its not worth losing terminal server failover for.

jkupski/compuveg-
It's not so much power spikes or under volts killing the occasional power supply (on a UPS) its more the massive dust bunnies that tend to get sucked into the PC's from the cooling fan's in the power supply's. I know, such thing as large dust bunnies shouldn't happen but they do, often... :\

compuveg-
Nothing as serious as "Hit button X *now* or make crater" but interesting and annoying enough that we get to play with Indemnification for some hardware. It was mostly that you lose a goodly block of clients when one switch goes on the blink for whatever reason. Having 1/2 to 1/4 of the clients go offline would just boost the chance that someone would lose a screen that they kinda needed at that time.