How much are deploy and capture speeds affect when FOG server is running from a Hyper-V VM?
-
Currently I’m running CentOS 8 with Latest FOG version and I got 6GB capture and 3GB deploy on images. Running from a Win Server 2012 Hyper V VM. Server is a Dell PowerEdge R815 32Cores and 64GB of RAM.
-
I have not personally setup a fog server under hyper-v. It would be interesting to know what your deploy time is to a modern target computer.
I can tell you on a well built FOG server running on ESXi VM under a pure 1GbE network you should get about (around) 6.2 GB/min. If your hypervisor is running a 10GbE core network you should see in the area of 13-15GbE to modern target hardware.
On the FOG server side disk subsystem and network subsystem performance is the controlling factors. CPU usage not so much. 2-4 vCPU should be sufficient. If you are going to have 1,000s of FOG clients running then you may need to increase the vCPU count.
-
@george1421 Deployment time to a Dell Latitude 3410 10th Gen i5 8GB ram and 128GB nvme. 9 minutes deployment time at around 2.5-3gb/min
Network is 1GbE switches and routers all around. I have it running on 8 vCPU the CentOS VM.
-
@tesparza said in How much are deploy and capture speeds affect when FOG server is running from a Hyper-V VM?:
I have it running on 8 vCPU the CentOS VM
How many computers have the FOG Client installed on them?
That number (GB/min) is lower than I would expect. Is that hyper-v running on bare metal or on top of a general windows install?
-
@george1421 Just 10 because I just set it up. Hyper-V is added role on Server 2012. The Server 2012 handle DHCP, I just added the role of Hyper-V and installed it.
-
@george1421 I had an Optiple 7010 wint Win7 that I had VirtualBox running CentOS7 with FOG and I was getting 7-8GB/min
-
@tesparza You might do simple network speed testing/comparison by mounting the FOG server’s NFS share from a fast client PC and copying files across to get an idea. From what I have read so far this is not caused by FOG and there is not too much we can do about it.
But still I might give you a hint on what to look at. You might take a look if the kernel used in CentOS 7 or 8 have the Hyper-V kernel options enabled. See details here: https://www.funtoo.org/HyperV_Kernel_Configuration
Not sure if you also want to install Hyper-V integration services (special drivers): https://www.microsoft.com/en-us/download/details.aspx?id=55106
-
@tesparza I have to be mindful to keep my hyper-v prejudices in check here.
The virtual box numbers are more in line with what I would expect for a well managed network. So the 3GB/min you are seeing is probably not a network itself issue.
So now we don’t know if its the hardware (R815 doubt), the disk or network subsystem, or running hyper-v in a type 2 hypervisor manner (I realize that hyper-v is more like a type 1.5 manner not type 1 and not really type 2). Would you get any performance gains by installing hyper-v on bare metal and then running your 2012 standard instance as a VM and your FOG server as a VM on that hyper-v host.
What type of disk system do you have in that R815 a H700 raid controller with raid-10 and 4 or more disks?
-
Ok I thought about this a bit more, it would be interesting to benchmark your FOG server running under hyper-v.
The rest of this post is based on the work I did here: https://forums.fogproject.org/topic/10459/can-you-make-fog-imaging-go-fast
So if you could setup a centos server on bare metal. To run as an iperf3 responder. Then from both the fog server and the 2012 run iperf3 or the windows equiv and test bandwidth between the external centos server and both windows and the fog server. That will give us a measurement of bandwidth loss between the native windows server and the VM client.
Then use the commands I used in the post to test (on the fog server) disk performance to get around any caching in the system you probably should use a 3GB file. That should flush out any cache between the fog server’s OS and the actual disk media.
Both of those tests will give us an idea how well the fog server is running under hyper-v. If I remember correctly the hardware I used for my testing was a Dell 790. So it wasn’t an optimized setup (done by design).
-
I ran FOG in Hyper-V. FOG being virtualized doesn’t have an impact alone. Resource contention might affect you if you over-provision your hosts. IOPS contention could affect you, network contention could affect you. So really comes down to the resources available to the hosts, and how much you can give to the FOG VM.
Personally, when I ran FOG inside of (insert any hypervisor here), the biggest two points of contention were network and disk. This is regarding imaging specifically. For non-imaging related tasks, like FOG Client related stuff, the biggest point of contention is absolutely CPU.
-
I don’t believe this is a Hyper-V issue. We have Fog running in a Hyper-V Ubuntu VM, with 8 CPU and 16g of ram, with no issues. The storage is backed by a hybrid disk array with 10TB of SSD accessed via ISCSI, and network is provided as 4x 40g links in a LACP bond.
Routinely see 13-15g on NVME backed clients, 4-6 on spinners, 3ish on the 5400 rpm spinners.
