I am not reaching the required speed

gusto

The variant 1 shows my real situation. If I measure the speed of the Internet, I achieve 500-600 Mbit

If I connect the computer directly to the modem (variant 2) I reach 800 Mbit.
Do you think there could be a problem with odroid H2 network cards (Realtek RTL8111G)?

pfSense 2.6.0-RELEASE (amd64)

stephenw10

Well yes the Realtek NICs don't help but you are probably limited by PPPoE there. Assuming you are passing the hardware through Proxmox.

https://docs.netgate.com/pfsense/en/latest/hardware/tune.html#pppoe-with-multi-queue-nics

Have you tried running pfSense bare metal on that?

Steve

gusto

@stephenw10
I didn't try bare metal because I don't have that option.
pfSense runs in KVM with 2 cores.

CPU Type	
Common KVM processor
2 CPUs: 1 package(s) x 2 core(s)
AES-NI CPU Crypto: No
QAT Crypto: No

Will it be better to use only 1 core?
I added a parameter net.isr.dispatch=deferred to System Tunable but nothing has changed.

JonathanLee

@gusto are you using a traffic shaping algorithm also just in case.

For example I have a 6meg DSL with traffic shaping it won't drop connections when it's overloaded.

gusto

@jonathanlee
I don't use traffic shaping

bingo600

@gusto said in I am not reaching the required speed:

I didn't try bare metal because I don't have that option.

I would support the request of running "Bare metal" , as it might be the combination of realtek + hypervisor latency that slows down the xfer.
By running bare metal, you would eliminate the hypervisor layer.

It seems that proxmox is based on Debian.
Wouldn't it be possible to boot from a "Live USB Stick" , and do the testing ?

/Bingo

Patch

@gusto said in I am not reaching the required speed:

Do you think there could be a problem

How have you configured the 2 NIC on Proxmox, in particular what else is using each NIC while testing?
Is the WAN and LAN traffic going through the same NIC while testing?
How are you testing the speed, through the Odroid H2 or to/from the Odroid H2?
With more NIC's using pass through for the NIC's pfsense uses may help.
depend in what pfsense and other VM are doing, the 1.5GHz Celeron J4105 may limit performance.

gusto

@bingo600
I really can't use bare metal. Proxmox runs a lot of lxc, which I can't turn off for a long time.
Yes Proxmox 7 is built on Debian 11.

@Patch
I installed pfSense according to the instructions.
NICs are burdened by other VMs and LXCs minimally.
I used to test the speed of the internet speedtest.net
All VMs and LXCs are almost always idle.
Measuring internet speed should not affect anything.

stephenw10

There's a good chance you're just at the limit of the hardware. It would not surprise me to see that sort of rate on PPPoE even if you were running bare metal. So with the VM overhead and whatever other loading you have you probably just need more powerful hardware.

Try running top -HaSP at the pfSense command line whilst you're testing. Of either core is hitting 100% you cannot go faster.

Steve

gusto

@stephenw10
During the measurement
top HaSP

last pid: 58660;  load averages:  0.67,  0.43,  0.27  up 9+13:13:59    22:40:44
172 threads:   3 running, 133 sleeping, 36 waiting
CPU 0:  0.1% user,  0.8% nice,  1.0% system,  0.9% interrupt, 97.2% idle
CPU 1:  0.1% user,  0.8% nice,  1.1% system,  0.5% interrupt, 97.5% idle
Mem: 26M Active, 79M Inact, 191M Wired, 97M Buf, 1638M Free
Swap: 512M Total, 512M Free

stephenw10

That's while it's passing 600Mbps?

That's hard to believe. Those numbers are what I'd expect when it's just at idle passing nothing.

gusto

@stephenw10
top -HaSP does not appear to show real-time values. Although I quickly clicked on the execute button, the values did not change much.
Today I did another speed test and saw a value of 700 Mbit.
CPU 79%

stephenw10

The output of top does update in real time. Just run it at the command line (ssh or console) then hit q to quit while the test is running to make it easier to copy/paste the values.

79% total CPU use is far closer to what I expect to see. And that's probably actually near 100% on one core which is the limit you're hitting.

Are you passing through the Realtek NICs in Proxmox?

Steve

gusto

@stephenw10
This is how I set up a network card for WAN (vmbr0).
It is similar for LAN (vmbr1).

f.meunier

@gusto
Hello
Are vmbr0 and vmbr1 attached to different physical NICs in proxmox ?

I had a performance issue on PCengine APU3 systems (looks very similar to odroid H2 ) : I could not get more than 600Mb/s
I swapped to ZOTAC Nano CI (or CA) systems and I managed to get approx. 900Mb/s raw throughput.

Patch

@gusto said in I am not reaching the required speed:

This is how I set up a network card

That is not pass through. You are using a virtual bridge and a virtual network interface to pfsense.

To use pass through implies you dedicate the physical NIC pfsense use to pfsense. See:
https://pve.proxmox.com/wiki/Pci_passthrough
https://www.thomas-krenn.com/en/wiki/Enable_Proxmox_PCIe_Passthrough

To do this you need more physical NIC than the 2 you have.

stephenw10

@patch said in I am not reaching the required speed:

To do this you need more physical NIC than the 2 you have.

...and it still may not help because you're running Realtek NICs and PPPoE. ~680Mbps is about what I expect from that.

Confirm it by running top during a test to see the per core CPU usage. The receive NIC queue is probably at 100%.

Steve