Hi,

Is it possible to have a UAG array (HA and NLB) across 2 physical locations? (e.g. 1 array between New York & LA)?

Thank you

What is the difference in latency between the two sites?

User from LA accessing NY resources and other way around?

Will have to find that out - what would be the cutoff point for this to work?

Hi,

Is it possible to have a UAG array (HA and NLB) across 2 physical locations? (e.g. 1 array between New York & LA)?

Thank you

Yes it is possible, but not with the built-in UAG clustering component as all servers need to reside on the same subnet for it to work.   Because of this limitation, we designed a UAG add-on software product that provides HA, NLB, and Geographic intelligent routing to UAG regardless of where in the world they're located.  Our solution is software based and runs directly on the UAG array master, and small agents on the members.

The geographic routing is quite interesting as our application retains a database containing all the regional IP information around the world and can send users to the most geographically appropriate UAG.  This is important because latency based routing isn't reliable any longer as most DNS servers will no longer allow external queries.  So, for your example you can setup geographic regions where everyone east of the Mississippi river go to your NY UAG and everyone west can go to your LA UAG.   You can even assign specific cities in states to a particular POP.  Our largest customer using this technology has 5 geographic sites with 70,000 users.

If you would like more information, just fill out the form and reference this TechNet posting:

http://www.portsys.com/index.php/contact-us

Will have to find that out - what would be the cutoff point for this to work?

Consider what performance your users are willing to accept, that will be the lowest. It depends on what your hosting behind the UAGs, if it is acceptable or not. In order words, even tho it might be technical possible, it might not be the right solution-

I will do some lab on the technical side of this during the next days.

thank you Keith - will take a look at the product you've recommended

No problem, I hear what you are saying - and unfortunately at this stage we are still collecting all the requirements - and mine was just a generic question.

thank you

Alright, 

So far no problems in creating an UAG Array over two networks, with a routing gateway between.

UAG 1: IP: 10.0.10.10 Subnet: 255.255.255.0 GW: 10.0.10.1

UAG 2: IP: 10.0.20.20 Subnet: 255.255.255.0 GW 10.0.20.1

DC: IP: 10.0.10.100 Subnet: 255.255.255.0 GW 10.0.10.1

GW: IP: 10.0.10.1, 10.0.20.1, DHCP Subnet: 255.255.255.0 

Now I need to test the functionality with some web servers also. I will let you know how it goes :)

If you want to create a real UAG array, the servers must reside in the same subnets, internal AND external. Microsoft will not support it otherwise. I know some companies have gotten away with stretched VLANs to have the same subnet IPs available in two different datacenters, but the last time I checked this is also not supported because the latency between can cause all kinds of issues.

Keith, is your solution just for UAG portal connections? Or does it help distribute DirectAccess connections as well? Just curious.

Hi Jordan,

It's for both DirectAccess and UAG Portal Trunks.     It took a long time to figure out the ISATAP stuff to get it working right, but it works quite well.

@Jordan
The only thing I agree on in your sentence is that latency will be the cause of your problems if you try to create an array over multiple networks. - But that is no different to any other cluster technology. The cluster needs to be able to communicate reliable between all nodes. What is defined as reliable, depends on the cluster technology. 

As far as I read the supported scenarios for UAG, its fully supported to create a array over multiple subnets, as long as your not using the underlying TMG as the routing device/gateway. 

UAG - Support boundaries
http://technet.microsoft.com/en-us/library/ee522953.aspx

However this is mostly theoretical possible only, nevertheless possible :)
I just don't like the statement saying it's not possible. When in fact it is possible. (I don't know the exact round-trip latency requirement for an UAG array, but just to compare, the maximum round-trip latency for a Windows Cluster across networks is 500ms.

Windows clustering and geographically separate sites
http://support.microsoft.com/kb/280743/da

I didn't say that it was not possible to do a stretched VLAN, only that it's generally not supported. This is information straight from the mouths of folks on the UAG support team. If you configure an array over a stretched VLAN and there are any kind of communication problems for which you want to open a ticket, you are likely to be asked to reconfigure your environment.

With a stretched VLAN we are still "sort of" talking about the same subnet. What you state about being able to do it on different subnets is defined in the documentation as not being supported:

All Forefront UAG servers that you want to join to an array must belong to the same subnet. - from http://technet.microsoft.com/en-us/library/dd857258.aspx

Jordan & Jesper:

I think you are both correct.  

Microsoft will only support an 8 server cluster (UAG/TMG clustering component) that all reside locally on the same physical subnet.  Stretched VLANs for the purpose of extending a subnet to create a UAG cluster is not supported as networking variables unrelated to the UAG product are introduced.

Using an external solution like the one I proposed earlier, or a hardware based load balancer solution (like Big IP) does not invalidate Microsoft support terms.  However it doesn't mean Microsoft will give you technical assistance with it, but it won't be an out of support scope configuration.    And to be honest, you're going to want support from the Load Balancer/Failover vendor anyway.

The only effective way to put UAG boxes across the globe in different geographic regions and provide failover, geographic intelligent routing, and load balancing is by going with a 3rd party solution.

he he, if it wasn't for the link in the email, I don't think I would have been able to find this thread again in the new forum layout.

thank you for all your efforts and comments on this topic!

Nice to here others doing the same thing as me... I am trying to implement a solution using UAG array and GSS accross the world (Middle east, Europe, Asia and US); however,  I get very odd issues when trying to sync (tmg shows sysned on some, UAG array member does not show synced, UAG manager shows member is synced :(

I bet this is why MS does not support it.  Hopefully I can get a few questions anwered here:

1. What the bandwidth requirment for array, what and how much data is sent between array members.

2. I did a netmon capture and saw dsadmin.exe connections every sec-- is this the array connection?

3.  When I ping the array memeber is returns a 220ms repsonse, is this ok for the array?

Thanks

@Z_Cory

1: First of, there is no requirements defined, as this is not supported at all. Meaning that if you ask anyone that does this for a living, it won't happen.

2: Properly easier to find out which port dsadmin.exe uses in order to identify the application. Also source and destination IP might be useful.

3: For default settings, yes its way to high. I'm pretty sure that it does not tolerate more than max 30ms or so between array members, with default settings. You might begin tweaking on settings for packet sizes, maximum wait etc, however that is entirely on your own.

Bottom line, as this might be theoretical possible with the current UAG application, I would definitely not go down that road. Based on all the input in this thread.