SG-3100 Slow Throughput

torred

@johnpoz said in SG-3100 Slow Throughput:

So directly connect you only see about 1/2 of what your paying for - and you don't think that is a problem?

I chatted with AT&T about this, it would seem that 600Mbit is "within tolerance" for their 1Gb service, smh.

I agree that these tweaks shouldn't make such a huge difference, so should I be contacting Netgate about a possible hardware issue?

johnpoz

what about the crappy 400 up? Is that in their tolerance range as well... They should advertise it as you "MIGHT" get close to gig with our service.. But prob not ;) If you get 400 then we say its good! So Piss off and send us your money!!

How exactly are you testing this? Take your isp device out of the equation... run say iperf..

iperf server --- wan pfsense lan --- iperf client..

What do you see then? This should be damn close to this

ifperf server ---- iperf client

If no pfsense and iperf serv and client are good, and with pfsense and tested same cables its BAD... like what your seeing then yeah I would be contacting support abut bad hardware..

What speeds were you seeing with it before you switched to ATT?

torred

So good idea removing the RG from the equation, here's what I came up with...

In summary, TCP looked "good" (~700Mbit - ~900Mbit), regardless of pfsense WAN or LAN ports used (also good if pfsense was not in the path, e.g. just across my switches). Sooo....hardware on the SG-3100 is good...hardware on RG is "good" (suspect)...but as soon as I connect the two they hate me.

I'm going to do some packet captures and see if I can gleam anything from those. If I find something interesting I'll reply back. If I don't reply in 3 days, I've thrown all of my equipment away and moved to the middle of Montana to start my life as a hermit.

Same cables used for all tests (Cat 5E)

--------------------- WAN-to-LAN tests (No ATT RG) -------------------

Laptop(iperf client)->(WAN port)pfsense (iperf server)
TCP: 737 Mbit/s
Reverse TCP: 809 MBit/s

Laptop(iperf client)->(WAN port-NAT rule)pfsense(LAN port 1)->[2x Netgear ProSafe Switches]->Internal server(iperf server)
TCP: 863 Mbit/s
Reverse TCP: 759 MBit/s

--------------------- LAN-to-LAN tests (No ATT RG) -------------------

Laptop(iperf client)->(LAN port 4)pfsense (iperf server)
TCP: 679 Mbit/s
Reverse TCP: 700 MBit/s

Laptop(iperf client)->(LAN port 4)pfsense(LAN port 1)->[2x Netgear ProSafe Switches]->Internal server(iperf server)
TCP: 909 Mbit/s
Reverse TCP: 795 Mbit/s

(Test without pfsense)
Laptop(iperf client)->Netgear Switch->Internal Server (iperf server)
TCP: 899 Mbit/s
Reverse TCP: 949 Mbit/s

--------------------- LAN-to-Internet tests -------------------

(SG-3100 and ATT RG)
Laptop(iperf client)->[2x Netgear ProSafe Switches]->(LAN port 1)pfsense(WAN)->(LAN Port 1)ATT RG->Internet VPS(iperf server)
TCP: 14.4 Mbit/s (this would be me uploading) -- Likely an issue with my VPS, it throttles uploads
Reverse TCP: 43.7 Mbit/s (this would be me downloading)

Same thing, except to http://speedtest.att.com/speedtest/
Upload: 47 MBit/s
Download: 76.6 MBit/s

(Just ATT RG, no SG-3100)
(I used the same cable that was between the SG-3100 and the RG)
Laptop(iperf client)->(LAN Port 1)ATT RG->Internet VPS(iperf server)
TCP: 14.8 Mbit/s (this would be me uploading) -- Likely an issue with my VPS, it throttles uploads
Reverse TCP: 21.6 - 455 Mbit/s (this would be me downloading) (Why the huge difference??? IDK.. sometimes low, sometimes 200s sometimes 400s... over 10 tests)

Same thing, except to http://speedtest.att.com/speedtest/
Upload: 124 - 459 MBit/s
Download: 268 - 828 MBit/s (Again, all over the place)

@johnpoz said in SG-3100 Slow Throughput:

What speeds were you seeing with it before you switched to ATT?

I've had AT&T for 3 years. When I got the SG-3100 back in October 2017 (as soon as it was released), my speed was good (600-700+). I've only noticed the slow down in the last month or two.

johnpoz

@torred said in SG-3100 Slow Throughput:

pfsense (iperf server)

That is going to show you low results compared to routing THRU pfsense.. You need 2 boxes.. Do not use pfsense as client or server in your iperf testing.

torred

@johnpoz I had only done that as one part of the test, as you can see, I did test through it in other tests.

I honestly do not know what is going on. I did a factory reset on the SG-3100 with the same results. Except now I'm experiencing a multitude of other failures.

I've removed my pfsense, and am just using the ATT RG and everything works perfectly.

Thanks for your help @johnpoz, I'll be contacting Netgate Support and see if they can help me out.

torred

Alright, if anyone happens to read all the way down here, I never figured out what the problem between the two was, but I ended up bypassing the AT&T RG by doing this:
https://github.com/aus/pfatt

It was fairly easy to compile the needed ng_etf.ko kernel module for armv6:

Get a FreeBSD 11.2 amd64 VM going (for pfSense 2.4.4) -- make sure to include src when installing
Get a shell, and do this:
$ cd /usr/src
$ make kernel-toolchain TARGET_ARCH=armv6
# Wait about an hour
$ make buildenv TARGET_ARCH=armv6 BUILDENV_SHELL=/bin/sh
$ cd /usr/src/sys/modules/netgraph/etf
$ make
You now have your own compile netgraph etf module, follow the rest of the guide.
I used the OPT1 (mvneta0) port for the RG, and the WAN (mvneta2) port for the ONT

Everything works fine now.

sean.allen

@torred - you're one step (but light years) ahead of where I am. I simply need a trusted copy of ng_etf.ko for FreeBSD 11.2 to plop onto my SG-3100 and I'm done. Everything else in the pfatt project is ready for the reboot.

I don't have spare hardware lying around so have been trying to download the FreeBSD VMware image, but it has no source. And when I try to download the source, it fails. This simple step mocks me. Any thoughts?

Sean

torred

@sean-allen, try following this: https://www.freebsd.org/doc/handbook/makeworld.html#updating-src-obtaining-src

TL;DR: svn update /usr/src

There's quite a few guides on setting up FreeBSD for qemu, virtualbox, and VMWare. Once you get it running it's pretty easy.

Edit: Also, you could...

wget ftp://ftp.freebsd.org/pub/FreeBSD/releases/amd64/11.2-RELEASE/src.txz
tar -xz -C / -f src.txz

sean.allen

@torred I'm very thankful for your help with the ng_etf.ko. I have all in place and am bypassing the AT&T RG.

However, my speeds have not improved. At all. Quite the let down as I was assuming the interaction between RG and SG-3100 was the issue

Directly through the RG I was seeing 900+Mbit up/down
SG-3100 through RG in IP Passthrough yielded ~100Mbit up/down
Same setup using PIA as my VPN gave ~75Mbit up/down
I bypassed the RG with great expectation and the ~100Mbit and ~75Mbit numbers remained. <sad trombone>

Those numbers varied, but not nearly as wildly as @torred results. The speed tests I'm doing are speedtest.net, dslreports.com and att.com. I'm not familiar with iperf. I loaded it on pfSense and the dizzying array of config options had me walk away from that.

Other than the PIA VPN, I have:

pfBlockerNG DNSBL
OpenVPN Server (though no clients, it's just there to hit my network from outside while traveling)
ntopng

Turning off DNSBL and ntopng have no measurable effect on speed tests. I have the laptop I'm running speed tests on directly connected to one of the switched ethernet ports on the back of the SG-3100 removing other switches from the test.

Any other thoughts or suggestions here? I feel like the SG-3100 should be able to keep up with these, even with VPN, based on what I've read. It surely should be going faster than it is.

Thank you!
Sean

Side note: Anyone know why I can't access these forums through my PIA VPN? I have to bypass that before any page will load.

Grimson

@sean-allen said in SG-3100 Slow Throughput:

Side note: Anyone know why I can't access these forums through my PIA VPN? I have to bypass that before any page will load.

https://forum.netgate.com/topic/136229/vpn-blocked

sean.allen

@grimson thanks! I searched the forum, but I kept getting assorted posts about PIA/VPN/access/etc. - none having to do with the forum.

Sean

sean.allen

It's clear that personal VPNs are a contentious issue here, so let's remove that from the equation for now.

I can get 900Mbit speeds directly from the AT&T RG, but as soon as I introduce my SG-3100 into the path (either through or bypassing the RG) I start getting 100Mbit (not through VPN) - or a bit more than 10% of the available bandwidth.

Any ideas on how I've messed up my configuration such that the SG-3100 is pouring molasses on my link? I'm going through my entire network to make sure I have "good" cables and switches to remove that from the equation - but even when I plug a new cable directly into the switched ports on the SG-3100, same result.

Sean

Derelict

The AT&T RG is a beast, unfortunately.

https://forum.netgate.com/topic/99190/att-uverse-rg-bypass-0-2-btc/1

sean.allen

@derelict Precisely why I went to such lengths to bypass it. Finally have that working (because @aus is a rockstar). Now the RG is not inline for standard (non EAP) traffic. It is SG-3100->ONT. That config yields less than 100Mbit, whereas if I do client->RG->ONT I get 900Mbit.

Derelict

It must still be something in the bypass configuration then. Perhaps something in the traffic that is not marked in some way that AT&T expects it.

Or something that should be negotiating gigabit is negotiating at 100.

If it were me - and I couldn't find someone else who has put all the pieces together - I would put a switch with a SPAN port between the RG and the ONT in this configuration client->RG->ONT and capture traffic on a mirror port.

Then I would put the same switch between the SG-3100 and the ONT in this configuration SG-3100->ONT and capture traffic and see if there is a difference in QoS bits, VLAN priority, or something.

johnpoz

Yup that is exactly the steps need to figure out what is going on

sean.allen

@Derelict and @johnpoz - thanks for the feedback!

Quick question, though: if client->RG->ONT is at 900Mbit and (SG-3100->RG->ONT or SG-3100->ONT) are both sub 100Mbit - doesn't that point to an issue with the config or hardware of the SG-3100? The RG running in IP Passthrough, or being bypassed, yields the same result when the SG-3100 originates the traffic. The bypass method would seem to not be adding or subtracting anything relevant here, but I defer to your expertise. The bypass, if you're curious, uses netgraph to set aside the EAP auth traffic such that it only goes between RG and ONT (which are plugged into the two routed eth ports of the 3100). All other traffic sent directly from the SG-3100 to the ONT via a new interface (ngeth0) defined by netgraph to tag outbound as VLAN 0 (some odd AT&T requirement). It would appear that the only thing the RG is used for by AT&T is to make sure AT&T equipment is present - so the hard-coded cert in the RG is required to authenticate the channel. Full details on the bypass, if interested, are here: https://github.com/aus/pfatt

The reason I ask is because it will not be easy for me to mirror and capture traffic as you've suggested. Partly because of hardware, partly because of expertise.

gsmornot

My SG-3100 runs in either bypass or IP-Passthrough with full speed. I would assume you have a config issue with the bypass setup. If you care to share screen shots I can look at what you have in comparison to what I have. I have been bypassed for better than a year now. I still put the gateway back from time to time to play with it but generally stay bypassed.

johnpoz

To do a span port all that is need is a 30$ smart switch and a box that can run wireshark..

So hardware constraint while you might not have on hand? What switch(es) are you currently using... And you do not have a laptop or pc - for that matter a current pi that you could sniff on?

You could even use your sg3100 as a sniff box for testing what is actually going on when your not using the sg3100 and seeing your 900mbps

Do you not have any smart switch?

sean.allen

@gsmornot - thank you for the offer! I am happy to share screen shots. Here or DM? What parts of my config are interesting for troubleshooting this? I used the bypass method described here. The only mods I made to pfatt.sh were to tell it which of my routed ethernet interfaces were connected to the RG versus ONT - and give it the MAC of my RG. Once it ran and did its thing, I configured the ngeth0 interface in pfSense to spoof the MAC of the RG. After that, traffic started flowing - just as slowly as through IP Passthrough :(

@johnpoz - this is my home network, I've just been using dumb switches. Netgear GS108, TP-Link (TL-SG1008D), etc. Yes, I can feel the impending mocking. I can go get a smart switch (recommendation?), no problem having a system to run wireshark. Interpreting the results is where I am quickly out of my depth. I wouldn't know what to look for. When my SG-3100 goes through the RG via "IP Passthrough" it sends supposedly unaltered traffic. When the RG is bypassed, it has been altered by netgraph, ostensibly just to tag it VLAN 0. Both paths out result in the same speed loss - so it doesn't make sense to me that it is something to do with the bypass config.

Is the theory that the SG-3100 does something to the packets, regardless how it makes it to the ONT, that slows things down? I did some experimenting with CoDel awhile ago because of massive bufferbloat on my former asymmetric link, but have since deleted everything under Traffic Shaping (because it didn't help anyway). Perhaps some remnant there? It's the only thing I can think of, but my scope of knowledge here barely scratches the surface of what y'all know.