2.2 on Hyper-V on Windows 8.1

MrGlasspoole

I also tried Gen 2 installation of pfSense and Debian and it did not work.
There is a table: http://technet.microsoft.com/en-us/library/dn848318.aspx

Beside that i build a machine for pfSense, FreeSwitch (Debian) and Win8 and it was not easy to find
a motherboard with the right NICs that support i3, i5 and i7. I wish there would be a board with
3 NICs but the ones out there just support i3 or Xeons.

Here is my build:
ASRock Rack Z97M WS
Core i5-4570S
8GB Crucial Ballistix Sport VLP
40GB Intel SSD 320
240GB Intel SSD 530
250Watt Delta Electronics DPS-250AB-53A from eBay for $10

Power usage is around 19~21W surfing the web.
My Asus Dark Knight router was using 10W so i think for what this machine can to
in comparison to the Asus the 20W are ok.

I'm still waiting for an answer from Intel on power consumption of PCIe NICs:
https://communities.intel.com/message/261778

Oh and i run all that on the 2012R2 Hyper-V Core :)

binaryjay

@hege:

did you disabled "Secure Boot"?
This is required for ubuntu, I think it will also be required for pfsense.

Yes, the error is the same… no boot device found. I get the impression that FreeBSD does support GEN2 but the pfSense installation ISO does not have the required UEFI boot files. If this is the case, it should probably be updated to allow for UEFI installations this day in age. I have not tried a UEFI install outside of virtual machine though.

tandem

My experience after 6 month Pfsense 2.2 on my HyperV-Homeserver under Windows 8.1 (6 clients with Windows and 3 Androids):
It works flawless.
Every reboot because of Windows updates starts Pfsense regular. Every update of Pfsense (every 2 weeks) works as it should.
I only had to restart my Fritzbox LTE three times because it was getting slow, but Pfsense switched reliable to the DSL WAN.
On my Homeserver I feel no difference in speed with or without Pfsense (HyperV shows 0% CPU).

binaryjay

@tandem:

My experience after 6 month Pfsense 2.2 on my HyperV-Homeserver under Windows 8.1 (6 clients with Windows and 3 Androids):
It works flawless.
Every reboot because of Windows updates starts Pfsense regular. Every update of Pfsense (every 2 weeks) works as it should.
I only had to restart my Fritzbox LTE three times because it was getting slow, but Pfsense switched reliable to the DSL WAN.
On my Homeserver I feel no difference in speed with or without Pfsense (HyperV shows 0% CPU).

I agree, it is working flawlessly for me on Hyper-V Windows 2012 R2 for me, I just wish the install would support Gen2 VM but it doesn't seem to matter much. It also doesn't report the status of the virtual NICs properly and in Hyper-V Manager they show as degraded but this is probably a FreeBSD thing. It's nice that I was finally able to retire a much slower Atom based box I had dedicated to routing to my server running on an i5-4590.

I was planning on having to move my dual NIC card to the server but as it turns out pfSense works fine sharing the LAN Virtual Switch with everybody else and I only have the second integrated NIC assigned solely to the pfSense machine with no host access. I did just move off of the pfSense DHCP to the Windows Server DHCP because I plan on one day playing around with the idea of running a domain instead of the workgroup… then I can rule over the family's internet access with godlike power.

binaryjay

Has anybody else noticed a pretty high system interrupts amount running pfSense on hyper-v?

Even just doing large file copy over the LAN (110MB/s) causes the server to use up about 10% of CPU time just on system interrupts as reported by Windows Server 2012 R2 task manager. Is there something I'm doing wrong with NIC/Virtual Switch setup? All of the offload features are enabled on the LAN NIC, VMQ is enabled. LAN NIC is even an Intel NIC. Seems a bit much to me… on a client with a similar intel NIC copying a large file from the file server to the client results in a less than 1% CPU time for interrupts.

tandem

@binaryjay:

Has anybody else noticed a pretty high system interrupts amount running pfSense on hyper-v?

Even just doing large file copy over the LAN (110MB/s) causes the server to use up about 10% of CPU time just on system interrupts as reported by Windows Server 2012 R2 task manager. Is there something I'm doing wrong with NIC/Virtual Switch setup? All of the offload features are enabled on the LAN NIC, VMQ is enabled. LAN NIC is even an Intel NIC. Seems a bit much to me… on a client with a similar intel NIC copying a large file from the file server to the client results in a less than 1% CPU time for interrupts.

I use my HyperV-Win8.1-Homeserver with a cheap dual Realtek-Nic, which is only connected with the Pfsense-Client for the DSL and LTE Wans and not connected to Windows (which uses only the local Intel nic).
The Homeserver is also doing the DVB-S Satellite connection with a dual link dvb-s-nic and recording and streaming it in my homenet with the software dvbviewer. Copying large video files produces nearly no CPU time.
The board I use is a Asrock Z87pro3 .

firewalluser

@maverick_slo:

@P3R:

@tandem:

Yes, HyperV on Server 2012r2 and Win 8.1 are a type 1 hypervisor…

This statement is a contradiction.

AFAIK the definition of a type 1 hypervisor is that it is the OS and runs on bare metal without any host OS. Therefore a type 1 hypervisor doesn't run ON Server 2012r2, Win 8.1 or any other host OS.

All I`m gonna say to this statement is a big fat LOL.

ESXi uses vmkernel for it`s OS. ESXi vmkernel IS NOT LINUX BASED.

http://www.v-front.de/2013/08/a-myth-busted-and-faq-esxi-is-not-based.html

binaryjay

@tandem:

@binaryjay:

Has anybody else noticed a pretty high system interrupts amount running pfSense on hyper-v?

Even just doing large file copy over the LAN (110MB/s) causes the server to use up about 10% of CPU time just on system interrupts as reported by Windows Server 2012 R2 task manager. Is there something I'm doing wrong with NIC/Virtual Switch setup? All of the offload features are enabled on the LAN NIC, VMQ is enabled. LAN NIC is even an Intel NIC. Seems a bit much to me… on a client with a similar intel NIC copying a large file from the file server to the client results in a less than 1% CPU time for interrupts.

I use my HyperV-Win8.1-Homeserver with a cheap dual Realtek-Nic, which is only connected with the Pfsense-Client for the DSL and LTE Wans and not connected to Windows (which uses only the local Intel nic).
The Homeserver is also doing the DVB-S Satellite connection with a dual link dvb-s-nic and recording and streaming it in my homenet with the software dvbviewer. Copying large video files produces nearly no CPU time.
The board I use is a Asrock Z87pro3 .

I have determined that the interrupts have nothing to do with the pfSense VM, it seems to be related to the Intel i217-V adapter itself and I can reproduce the issue coping files over the network to the hyper-v host even with the pfSense machine off.

DPC latencymon software points the finger at ndis.sys (no surprise), can't seem to get anywhere past this… the NICs on my host are the intel i217-V (LAN) and Atheros AR8161 (WAN) but who knows the issue could be the hyper-v virtual nic on the host (which the host shares with the virtual switch).

I dunno, it's really not THAT bad, but it can occupy up to 8-10% of CPU time on a 110MB/s file transfer which is ridiculous for an i5-4590... practically it won't really matter that much for my use but it just bugs the hell out of me. Aready tried tweaking around with all of the advanced nic settings with no difference other than being able to make the interrupt CPU time even higher by disabling interrupt moderation completely.

tandem

My board has also a Intel i217-V nic.

binaryjay

@tandem:

My board has also a Intel i217-V nic.

Can you share what driver version it is running and what advanced settings you have set for the driver? Have you tried coping a file over the LAN directly to your hyper-v host and watching task manager on the host as you do it? Perhaps it's the Atheros adapter after all but I don't see how given that I only have that adapter mapped to the pfSense WAN and I get the high interrupts just copying a file over the LAN to the host… I really don't want to add another NIC to the thing since it's an ITX build I put together for lowest power consumption possible and chose the motherboard specifically because it had dual LAN built in. I don't want any more unnecessary power drain nor do I want to give up the one and only PCIe slot for another NIC.

Scratch that....

You know what, it could be that task manager is just useless in a hyper-v environment... I decided to run cpu-z while doing the performance testing of the network. Even though task manager reports a good 8% CPU use from interrupts or so when doing a file transfer I noticed cpu-z is reporting that in this situation my CPU is not even leaving it's lowest power state and frequency (it is running at < 800Mhz at this time). Task Manager itself running on the host reports the CPU locked at it's "rated speed" but we know that this is not the case.

To test my theory I artificially loaded the CPU by running prime95 in the host and then running the file transfer test and behold that if I load the CPU in order to make it run at it's full rated speed the CPU use numbers from system interrupts during the file transfer decrease dramatically to only between 1 and 2%.

Somehow, I always assumed that the windows task manager was smart enough to report the CPU load based on it's rated capacity and not based on it's current power state. Today I learned this is not the case. In fact, my network only consumes 6-8% of the CPU maxing out the full gigabit connection rate while in the lowest power state of less than 800mhz, in other words it doesn't even break a sweat. In fact, I had a hard time finding out a way to coax it into a higher power state and just threw the kitchen sink at it with prime95 and that's with the host already running 3 VMs including a second full instance of Windows Server 2012 R2 as well as with XBMC running 24/7.

I have to say, I am actually very impressed with this haswell part (i5-4590) now considering it must be doing all of this using next to no electricity normally.

P.S. I ran this test with all offloading features enabled and receive side scaling maxed out. In my case the host shares the virtual switch with all of the VMs LANs. Case closed!!!