I have a 2012 R2 SP1 primary site that is servicing the main campus of the University I work at.  All the site servers are located in the main campus data center.  We also have a remote campus about 15 miles away from the main campus which utilizes the the servers in the main campus data center.  For the most part this hasn't caused any issues, however when techs at the remote campus try to PXE boot and image a device, the TFTP portion of the boot process takes 10+ min to download the boot image as opposed to the 30 sec it takes on the main campus.  Compounded when imaging multiple machines at once, 10 minutes turns into 20, 30, and so on, and is not feasible for the techs at the remote site.

 

We have worked with our network engineers to verify that there were no problems on the network causing this difference and after A LOT of testing we determined that everything is working correctly as it is currently designed.  The time difference comes from how TFTP works with the whole send 1 packet, receive 1 packet process.  On the main campus, this isn't much of a problem but the minuscule bit of extra time between packets going back and forth from the main campus to the remote campus adds up to the extra time in the boot process (we actually drew out the math).

 

So now, half of us are of the mind that we need to put a DP at the remote campus wants to start doing registry hacks and messing with DLLs to increase the TFTP window size.

 

Is there any criteria (Physical distance, bandwidth, latency, clients managed, etc) on when it is appropriate to place a DP (or any other roles) at a remote site?   Any documentation I can show about the matter would be helpful.  Thanks!

Well DP can support 4k clients and i think around 10k packages/application so the limitation are there.

For the network part i think you already answered yourself when you said that the time it take is so much longer due to the link.

Normally you put DP to avoid people from using link between the site. The closer the clients are to the DP the faster the download will be. 

You see remote DP if the company as some sort of network block that prevent client from moving over certain boundaries.

For the other roles the MP can have up to 25k clients or so.

No clue what other roles you are using.

https://technet.microsoft.com/en-us/library/gg682077.aspx?f=255&MSPPError=-2147217396#BKMK_SupConfigClientNumbers

• Edited by Frederick Dicaire 15 hours 51 minutes ago

Well DP can support 4k clients and i think around 10k packages/application so the limitation are there.

For the network part i think you already answered yourself when you said that the time it take is so much longer due to the link.

Normally you put DP to avoid people from using link between the site. The closer the clients are to the DP the faster the download will be. 

You see remote DP if the company as some sort of network block that prevent client from moving over certain boundaries.

For the other roles the MP can have up to 25k clients or so.

No clue what other roles you are using.

https://technet.microsoft.com/en-us/library/gg682077.aspx?f=255&MSPPError=-2147217396#BKMK_SupConfigClientNumbers

• Edited by Frederick Dicaire Friday, August 28, 2015 3:59 PM

Is there any criteria (Physical distance, bandwidth, latency, clients managed, etc) on when it is appropriate to place a DP (or any other roles) at a remote site?   Any documentation I can show about the matter would be helpful.  Thanks!

You won't find any hard numbers saying to place a DP if you have X bandwidth or Y latency. As Frederick mentioned, a DP can manage 4000 clients. It also depends on the number of deployments you're making.

The critical factor is the network link between the 2 sites. Just by saying that your boot image takes 10 minutes to download would be a good indicator for me. How many time does the whole OSD process takes ? Maybe 2-3 hours for a simple deployment ? It just doesn't make any sense and that's excluding "regular" deployments to the deployed client in this site.

If you have a couple hundred remote clients and you're deploying a images, managing Software update and make software deployment over a slow high latency link, you should have a local DP.

https://technet.microsoft.com/en-us/library/gg712321.aspx?f=255&MSPPError=-2147217396#BKMK_DistributionPointInfrastructure

The question here isn't really about DP placement for the delivery of content, it's about DP placement for the purpose of providing PXE boot services. All of the normal guidance given (including that given by Frederick and Benoit) thus doesn't really apply.

Basically though it comes done to how much of a pain is this for you? It sounds like it's a pretty big pain and if placing a DP at that location addresses this pain, I think that's your guidance right there. The only cost is the server -- which can be just a pizza ox or even a desktop class system and a Windows Server license (along with the normal support costs also).

This sounds like like a no-brainer to me, place a DP. Not sure how the documentation here helps at all or why you are hesitant. You have an issue that would go away by simply adding the DP; q.e.d, add the DP.

...and if it doesn't work or help the way you wanted it to, then just remove it. Pretty simple really.

I agree that it's an obvious choice to place a DP at the site, however i'm not the one that is writing the check for the hardware and storage to do it.  The people who are are the ones looking at it and saying "we have a 10gb fast link to that site so if PXE is slow there must be something not working correctly" even though we have ruled out any network issues.  I was hoping there was some documentation or recommendations that I can present to show that in cases like this, a DP at the remote site is the best solution.

 

You can easily test it with a desktop PC with Windows Server installed. Minimal (if any) hardware cost for a Proof of Concept. Even a desktop machine with a single HDD can handle doing several machines PxE booting off it at once so it should give you plenty for any test you need to confirm if it will work.

There is no documentation that says "if this then that" as there are far too many variables to make it that straight forward, but if you do have 10Gb links then there is definately something else wrong. We do that comfortably across WAN links between data centres with 1Gb links. Perhaps check there isn't QoS or some other shaping going on?

There are some tweaks you can do with Pxe TFTP packet size that might help.

https://kickthatcomputer.wordpress.com/2013/05/30/how-to-boost-pxe-tftp-boot-speed/

• Edited by Scott Williams - IP1 15 hours 58 minutes ago

How big is your boot image? How long does it take to copy a file of that size from Site A to Site B? Does that also take 10 minutes? Is that reasonable?