Ideal Fog Server Setup



  • Good Evening All,

    I have been using fog in my school district for about a year now, things have been great and most issues I have been able to locate on the forums and correct. However I have a couple of nit picky things that I would like to know more and learn how to correct in the future if I am able to with not much $$$.

    Currently we have about 8 buildings, 6 of which are all fiber connected at 1GB (backbone), we have 10/100 to desktops, 1GB to servers. Our Switches mostly consist of 8 year old Procurve 2650s, and 5308xls. When we image Labs we have a standby 24x 100/1000 Switch for imaging labs with Multicast and it runs great, we plug in Fog, plug in as many cords as we can and we can blow a 30GB image out in under 8 minutes… For Classrooms, we do Unicast with a Queue and the first 2ish machines go good, and any other start slowing down… I will explain my server setup next.

    We have a Fog Server and 1x Storage node. as I said, when we do Labs (30) or so in one room, we plug all them hard into our 1GB switch and it works GREAT…

    Then we have a Fog Server SPECD as
    (HP DC8000 Elite, 2GB Ram, 1x 600GB 10K velociraptor, Duo Core @ about 2.8Ghz)
    and 1x Fog Storage Node, HP DL 380 G5, in a Raid 5 Setup with 36GB drives I think they are. Memory is 4GB with Xeon processors about 2.4-2.8ghz. We have the Unicast set to 3 and 3 on both servers to handle clients… The first two start great on Unicast around 15 minutes or so for 25GB image, the other 4 can take about 30-45 minutes and then they re-queue when spots are open. The main problem we see if when we have 3-4 going, the Kernel Download is ridiculously slow and Ive read on here some people can Unicast like 20 or so at a time when decent speeds. This mostly leads me to believe I have a bottle neck in one of 2 points.

    Hard Drive (single drive and one storage node)
    OR
    1GB Nic choking on 100 Port (when we have it deployed in a building). At home base the server has 1GB connection but when we take it to a building to image labs, we let it sit in the building so it becomes 10/100 speed and image the building from with-in the building.

    We don’t have anything VLANd which is something we are going to fix when we get new Networking Gear, and we will most likely end up with 10GB Backbone, and 1GB to all desktops in 1-2 years. So it could also be the Network causing slow Unicast, but I think im more inclined to think the NIC or Hard disc, and if so what would be the proper remedy? Get another NIC and Bond them? do I get another hard drive and Mirror them? do I Raid 5? What is everyone else doing and what has worked for you?

    Thank you all for any advice,

    Kyle



  • [quote=“Kylejb007, post: 4521, member: 1442”]Chad and Raff,

    That Did it!!! The intial results have been really impressive. Today we imaged our high School (Classrooms), which is approx. 80 machines… Normally we use to use Novell Zenworks to image them and it would take practically 1 whole day to image (at least 4-6 hours) and we would come in the next day and join them to the domain. Well im pleased to say that we started this morning at 9am, and completed the last computer (we dont have the fog service setup yet) we joined the last computer to the domain at 3:30PM. All of the machines loaded up fine with no Kernel downloading slow, and the Storage Node happily pumped images out to 10 machines at a time with everyone taking roughly 25-30 minutes to complete each. We even had to re-queue a couple computers due to Sysprep failing to process the unattend file, I assume this happens due to a network glitch pushing it to it or hardware glitch on the end station, usually a second image to it and it works rarely have I had to do it a thrid time, I end up switching patch cords and that fixes it.

    Do I dare increase the amount of machines imaging at one time or I do have another decomissioned G5 I could use if I wanted… This is our largest building so the others shouldnt have any issues at all with time, I just figured for planning in the future to do 20 at a time would speed things up tremendously.

    Definitly thanks for the advice, best product around![/quote]

    Same experience here. We use ZENWorks for other functions, but dropped it for imaging purposes as soon as we found FOG!


  • Moderator

    Get the fog service installed on your images and you can skip the joining to the domain step (with some additional settings in Fog Web UI).

    I would increase the queue until it starts affecting the imaging times of the machine by more than a few minutes. It’s really subjective whether you want to do 10 computers every 25 minutes, or 15 computers every 38 minutes.

    With a good disk subsystem and raid controller, you can push the server to it’s limits.

    For my setup, unicasting is actually faster than multicasting, and much more reliable.



  • Chad and Raff,

    That Did it!!! The intial results have been really impressive. Today we imaged our high School (Classrooms), which is approx. 80 machines… Normally we use to use Novell Zenworks to image them and it would take practically 1 whole day to image (at least 4-6 hours) and we would come in the next day and join them to the domain. Well im pleased to say that we started this morning at 9am, and completed the last computer (we dont have the fog service setup yet) we joined the last computer to the domain at 3:30PM. All of the machines loaded up fine with no Kernel downloading slow, and the Storage Node happily pumped images out to 10 machines at a time with everyone taking roughly 25-30 minutes to complete each. We even had to re-queue a couple computers due to Sysprep failing to process the unattend file, I assume this happens due to a network glitch pushing it to it or hardware glitch on the end station, usually a second image to it and it works rarely have I had to do it a thrid time, I end up switching patch cords and that fixes it.

    Do I dare increase the amount of machines imaging at one time or I do have another decomissioned G5 I could use if I wanted… This is our largest building so the others shouldnt have any issues at all with time, I just figured for planning in the future to do 20 at a time would speed things up tremendously.

    Definitly thanks for the advice, best product around!



  • The DL 380 G5 has a Raid 5 of 8x 72GB, 15k RPM SAS drives. (verified it just now)
    Both Machines have 4GB of ram/

    I went into my Fog Setup and I set my Main Fog Server to

    Max Clients: 1
    and the Storage Node: 10

    I will test this next week most likely is when we will go to the next building to start cleaning/imaging.


  • Moderator

    The intarweb ate my reply earlier, so I hope it enjoys having indigestion.

    For mainly unicasting, set the storage node (dl380) up to have the majority of the clients, since it has a faster disk subsystem (I’m assuming it’s got 4-6 drives in raid 5). Go into the storage node settings and change the defaultMember (should be your HP machine) to have a queue size of say 1, and change the other storage node (the DL380) to have a queue size of 10.

    For mainly multicasting, the computer with the most RAM would probably be the best to use for the source of your images.



  • Raff,

    Thanks for the information, so if I understand the “Master Node” setting, What Host I decided to make Master will handel all most imaging functions except Kernel Download ( does it also Include being the host for Multicast Sessions)? If so I may want to change it up, the DL 380 is racked and mounted and my Fog Server is the DC8000, it would be ideal for me to have my Rack Mounted server be the Main host that carrys the hosts (mysqL), and distributed the Kernel and make my DC8000 the Master Node? (That way I can still take the Node with me Onsite and Multicast, and then bang out the computers… Or at this point to remedy it, could I set my DC8000 to Image 0 Clients and make the Storage node take them all on? Would that work to for now?



  • Kyle,

    You can solve your speed issues quite easily. Don’t image from the HP DC8000, use this for only MYSQL, FOG CONSOLE,TFTP, PXE only (the kernal transfer is slow because its coming from the HP DC8000 and the imaging process it eating all its bandwidth).

    The HP DL 380 should be a master storage node, and imaging should be done from this node or any additional master nodes you add in the future.

    Your setup should then be able to handle about 10-12 simultaneuos unicasts at each master storage node before you notice a drop in performance for that node only.


Log in to reply
 

944
Online

39.3k
Users

11.0k
Topics

104.6k
Posts

Looks like your connection to FOG Project was lost, please wait while we try to reconnect.