Nethserver vs pfsense: QoS

(dreamerman) #1

Hello, I am on NS v7.4 and currently have a working pfsense box. When I quickly look at NS’s QoS, I thought it is not as “strong” as pfsense. I do not know enough to compare each function side by side but just a feeling when I skim the doc. Can someone tell me if NS QoS is better/worse/same as pfsense?

(Filippo Carletti) #2

I can speak for NethServer.
We don’t shape (i.e. limit) traffic, we prioritize it trying to always use all the available bandwidth.
We use a vanilla shorewall simple traffic shaping setup:

I would like to move to shorewall complex shaping to take advantage of codel (but I have a patch for codel on simple shaping). I also have a test setup with complex shaping and hfsc.

To take an informed decision I would like to make comprehensive tests. If somebody here could run Flent ( on a variety of internet links I will provide test material.

(Jeroen Visser) #3

Any time … but I can and will not use nethserver as firewall due to how that would impact our production.

I would need a maintenance weekend to use nethserver for those tests. Read; not any time soon.

I DO have 2 pfsense appliances running … with an ipsec site to site between them, an openvpn road warrior setup and a bunch of site to sites with customers.

I have no issue stress testing that in off hours :slight_smile:

(Filippo Carletti) #4

If you want to test a more advanced traffic shaping using codel ( follow this steps:

yum --enablerepo=nethserver-testing update nethserver-firewall-base
config setprop firewall tc Internal
signal-event firewall-adjust

Look for code in shorewall show tc output to confirm activation. Example output:

#shorewall show tc | grep codel
qdisc fq_codel 3: parent 1:11 limit 10240p flows 1024 quantum 1514 target 5.0ms interval 100.0ms ecn 
qdisc fq_codel 4: parent 1:12 limit 10240p flows 1024 quantum 1514 target 5.0ms interval 100.0ms ecn 
qdisc fq_codel 5: parent 1:13 limit 10240p flows 1024 quantum 1514 target 5.0ms interval 100.0ms ecn

I suggest using flent on ubuntu following instructions. Please test buffer bloat before and after update.

@alefattorini, do you think that this deserves a separate topic?

(Alessio Fattorini) #5

Nope, I think that it’s ok here.

(Rob Bosch) #6

You can always opt for a dedicated Gateway role on a separate NS instance. Then your other services are not affected by NS gateway service.

(Jeroen Visser) #7

True, but still be behind a pfsense firewall … and that would not work too well in a comparison.

(Pablo Sanchez) #8


Are you still looking for testers?

I tried following the recipe but “yum” found no packages to update.

These are my current version:

# rpm -qa|fgrep nethserver-firewall-base


(Filippo Carletti) #9

Yes, I’m still looking for testers. I’m working on yet another setup, I will update this topic when everything is ready for testing (next week, I’m on vacation now).

(Pablo Sanchez) #10

Hi Filippo,

I hope you enjoyed your vacation.

Do things still look like you might have some bits to be tested this week or would it be for next week?

My environment:

  • 120 Mbps down / 20 mbps up
  • Five homes sharing the Internet
  • Slight limitation, some of our wireless backbone is only 100 Mbps
  • We use Zabbix to monitor ICMP vs WAN Throughput


(Filippo Carletti) #11

I still have some problems with mark-based traffic classification. I need some time, sorry.

(Pablo Sanchez) #12

No need to apologize, you are help us! :slight_smile:

I’ll stand by until I hear from you.

Thank you for working on this feature. I appreciate it!

(Pablo Sanchez) #13

Hi Filippo,

Have you had a chance to look into the mark-based traffic classification or is it a tough problem to solve?

As I don’t want to be a pest, would you prefer I just wait for an update to the thread or to periodically ping you. :wink:

Thank you!

(Filippo Carletti) #14

I have completed my tests over the weekend.
It has been an amusing full immersion experience.
The good news is that I achieved all targets.
The bad news is that we need RHEL 7.5, so the release is delayed for at least a couple of months, probably a bit more.

On the other hand, I’m using an improved QoS configuration in production and it seems to work better than the standard one.
I plan to ask for merging of this improved version
Would you like to test it using 2 custom templates?

(Pablo Sanchez) #15

Hi Flippo,

Is RHEL 7.5 truly needed or can CentOS be used?

I’d love to test the two custom templates but CentOS 7.5 is not released. Hmmm. What do I need to do in order to test them?

Thank you!

(Filippo Carletti) #16

CentOS 7.5 will be fine (when released), but it’s not yet available, so I had to use RHEL 7.5 beta1 for the tests.

The custom templates will only enable fq_codel which behaves better than sfq.
Give me some hours to create a package.

(Pablo Sanchez) #17

It sounds as if I will have to patch the O/S on my NethServer 7.4.1708 eh?


(Filippo Carletti) #18

I have built updated packages.
Temporary instructions:

yum update
config setprop firewall tc Internal
signal-event firewall-adjust

I suggest setting ingress (inbound/download) bandwidth to empty or zero. If you put a limit on ingress (why?), you may need to disable advanced features on some ethernet cards.

ethtool -K $IF tso off
ethtool -K $IF gso off

Substitute $IF with the wan (red) interface name.

(Pablo Sanchez) #19

Hi Filippo,

I am currently running the latest bits. Thank you!

I have some data.


If I clear out outbound bandwidth (blank) as well, I get an error[1]. The gateway also fails to forward packets.

If I reset the outbound bandwidth, the gateway resumes forwarding packets.

Stress Test

I ran a stress test where I downloaded an ISO at full speed. In parallel, I made a VoIP call. I didn’t run a full-on ping test but I captured some ping data[2] - subsequent post. The ping is to my ISP’s gateway.

The results of the test were promising but not perfect. :wink:

The phone conversation was noticeably affected by the download. When I paused the download, all was well. Basically I could hear the other party but there were times when the call was a bit choppy.


[1] - setting outbound bandwidth to (blank) too:

(Pablo Sanchez) #20

Addendum … as I’m considered a New User I can only post one image per post.


[2] - WAN vs ping time (to ISP’s gateway) during soak test