One of the reasons I went with No One was because I don't want an Amazon owned router controlling my network, and the DGA4134 they supply seems to review well enough, doesn't require a mobile app to set up, and doesn't hide features from me or paywall stuff.
City Fibre
- Thread starter PzR Slim
- Start date
Associate
- Joined
- 2 Feb 2008
- Posts
- 1,914
- Location
- Rainham - Kent
I did try the eero Pro6e router supplied and it seemed fine for my usage which is just home user / gamer, but I already had a Asus 2.5g compatible router so just used that and left the eero in the box, the tests I used with it to see if it was better or worse than my Asus RT-AX86U all came out the same apart from wifi which didn't seem to have such good range in my house, but I didn't try tweaking it.
Last edited:
Just had No One installed. Appointment was scheduled for PM (13:00 - 18:00) City Fibre guys were great. Clean install. Had to wait for a MEWP to run the fibre from the pole, but they arrived at 13:15 and was gone at 14:30.
Went for 900/900 for £38.99. Using the supplied router at the moment, but will get my OPNSesne box setup the weekend.
hi, i was wondering if you managed to get opnsense router working on your noon/cityfibre connection successfully? I'm on cityfibre network but with vodafone (900mb package) and have been struggling to get 3rd party routers working at full speed. so far i've tried an asus ac87u, a draytek 2865ac which only seems to max downloads around 500-600mbs. uploads tests no problem getting full 900mbs. i was told it could be lack of cpu power due to pppoe which takes a lot of grunt. now I'm trying a fanless mini pc box with opnsense and has an intel n100 which is quad core 3.4ghz and should work but this seems to be maxing at downloads of circa 700mbs. i was just wondering if you have any feedback with the opnsense. thanks
Hi, I do now run an OPNsense box to drive my connection. My setup is an old HP Elitedesk 800SFF G2 which has 8gb DDR4 and an i3 6100. NIC's are onboard Intel i219LM which I use for the WAN interface and a Intel T340 4 port for the VLANS I run. This sits at 15w when idling, jumping to 20w when pushing the connection. I can get the full speed of the connection without the box breaking a sweat. I haven't done any tuning or anything. All the Hardware accelerations are off as they can introduce issues depending on NIC's/Drivers. I would have thought the n100 could push the connection easy.
Here was a download from steam the other day. Had to laugh
Last edited:
Hi, I do now run an OPNsense box to drive my connection. My setup is an old HP Elitedesk 800SFF G2 which has 8gb DDR4 and an i3 6100. NIC's are onboard Intel i219LM which I use for the WAN interface and a Intel T340 4 port for the VLANS I run. This sits at 15w when idling, jumping to 20w when pushing the connection. I can get the full speed of the connection without the box breaking a sweat. I haven't done any tuning or anything. All the Hardware accelerations are off as they can introduce issues depending on NIC's/Drivers. I would have thought the n100 could push the connection easy.
Here was a download from steam the other day. Had to laugh
thanks for sharing that. i'm really at a loss why 3rd party routers are not maxing out the 900mb download whereas the vodafone supplied router does it with ease.
the pppoe settings are the credentials supplied by VF with VLAN tag to 911 being Cityfibre. adding an MTU setting of 1492 and also cloning the vf router mac into opnsense makes no difference.
are there any other settings which you have for you the pppoe wan that i may have missed?
I can't see anything you've missed. What NIC's are on the N100 box? Intel i226v's?
Here are my settings
I can't think of anything else. You testing with the same speedtest?
FreeBSD w/ Intel NICs have a bit of a limitation when it comes to PPPoE in that it can only be processed by one core (RSS basically only balances traffic to one queue, which can only be processed by one CPU core). There are some tweaks to get around this for PFSense (probably the same for opnsense) but I believe that doesn't fully resolve the issue, either get yourself a Broadcom NIC (like a BCM5719) for cheap, or maybe try something like OpenWRT?
I can't see anything you've missed. What NIC's are on the N100 box? Intel i226v's?
Here are my settings
I can't think of anything else. You testing with the same speedtest?
yeah your settings look like same as mine. i'm not sure if there is a particular credential that vodafone are detecting for their router that makes it full bandwidth. weirdly when i spoke to them earlier in the year about some speed issues, they where able to identify the serial number of the supplied router and also do a remote update so ames me wondering if theres something in the background.
i heard about the single core issue with freebsd but when looking this up, it's seems to show that this was resolved recently. i may give open wrt a try just to rule it out though.
Had the same issue with OPNSense (FreeBSD), so went with MikroTik CHR (Linux) instead, you could also go with OpenWRT (Linux).
There's a lot mentioned about it:
https://www.google.com/search?q=freebsd+pppoe+perfromance
https://forum.opnsense.org/index.php?topic=27049.0
With MikroTik CHR easily get 940/940 with Vodafone using a i7 920, Intel I350 NIC.
Not sure how much you're able to tune the settings, you might be able to find some recommendation by searching through what I linked
Last edited:
@kungfuman what hardware are you using?
The latest router is a fanless mini PC with an Intel N100 CPU, 8gb RAM with I226-V NICs. This is currently maxing out around 700mb on speedtest.net. Plugging the Vodafone router back in give me full 900mb/s (probably 930mb/s most tests).
So far, I've also tried:
ASUS RT-AC87U which was testing around the 600mb/s
Draytek 2865AC which was also around 600mb/s mark with HW acceleration turned on.
Uploads doesnt appear to be a problem for all the units and all get a max upload speed of 900mb/s
The credentials I am entering for each unit are the PPPOE log in details provided by VF and also adding a VLAN tag of 911 to the WAN which is whats suppose to be for Cityfibre network.
From a 'functional' point of view, the issue is down to the Intel NIC hardware.
Basically, when a packet arrives on the RX ring (inbound into the network port / downstream or download), each packet is assigned to a queue based on a hashing algorithm. The problem with PPPoE is there isn't an 'entropy' factor the NIC can use to assign it across multiple queues (because it can't look further in the packet, it's just a PPP encapsulated ethernet frame), so the traffic always ends up in a single queue and can only be processed by a single core by default for all the actions (enqueue, decapsulation, NAT, routing lookup etc).
There are tweaks like setting net.isr.dispatch to deferred, along with net.isr.bindthreads and net.isr.maxthreads which don't fix the problem with the IGB NIC hashing mechanism but instead change the allow that single queue and actions to be processed by multiple cores/threads, rather than a single core.
I did a little lab testing with PFsense Plus and a Vyos router running a PPPOE server to see what sort of throughput i could get, both were running on the same Proxmox box with 4 vCPUs to PFSense and 2 vCPUs to Vyos (i7-10700K) connected over virtio interfaces. With basic iperf3 testing, I managed about 3.26Gbps bi-dir before I started seeing significant loss, so it does improve things.
The long and short is though, that modern intel NICs lack the ability by default to balance inbound PPPOE to multiple RX queues, whos interrupts are generally by default configured to be processed by multiple processors, if you move over to a BCM5719 NIC or an older Intel NIC, you'll get around the problem.
Also, if you're running a CPU with hyperthreading, make sure you disable it. It hurts more than it helps.
Last edited:
The CPU usage is still low when testing, between 10-12%. I havent tested with multiple computers yet as I have only connected my laptop to the single LAN connection on the router. I could try testing with multiple speedtest using a switch later tonight and see if there's improvement.
I used a similar micro/mini-PC or whatever you call them, with Untangle and also OPNsense and I was able to reach over 900 Mbps.
If it's from China, take the CPU heatsink off and re-do the paste. The one I had and sold to a mate of mine, he said the thermal paste was gash and it significantly dropped the temperatures.
It doesn't explain why the connection isn't performing with the ASUS and Draytek.
If it's from China, take the CPU heatsink off and re-do the paste. The one I had and sold to a mate of mine, he said the thermal paste was gash and it significantly dropped the temperatures.
It doesn't explain why the connection isn't performing with the ASUS and Draytek.
Yes I think I will give that a try in the coming days just to test it.
I am wondering if this is due to the vodafone side of things. I had some major speed issues over a course of a few weeks late last year (was only get 20-30mb/s down at one point) and VF advise me to wait and they should resolve themselves. They did eventually resolve but I was still only getting 500mb/s max for a while. During a call to their tech support, they were able to log into the VF supplied router remotely and supposedly made a configuration change and then normal speeds (900mb/s) came back. I am wondering if this 'backdoor' config change is whats restricting the speeds unless the VF router is used.
You have to remember depending on where you're looking it's not an exact measurement and is the overall CPU load not per core, best way to check is to log into the CLI and run 'top -P' and look at the load on each core when you're running a download / speedtest, look at what it peaks at.
Also, it's not just about raw clock speed / usage, if you have a single core enqueuing and processing traffic, you simply cannot push the pps needed through it because it has to process one frame before it picks up the next off the RX ring, this is a buffer so it will pick it up and process it FIFO by default, and once the buffer gets full it will taildrop and you will see your TCP performance drop.
If you look at the blue line on speedtest.net when you run it, is the line flat, or does it have a sawtooth effect (hills and troughs)?
If you get 900Mbps with the supplied router and not with your own x86, I guarantee you it's down to the settings / setup of the x86 device. Try setting these tunables settings and see if you see a speed improvement first and foremost (will be honest here, i've only done this on pfsense not opnsense, but it's a freebsd setting so it should be appropriate).
net.isr.bindthreads: 1
net.isr.maxthreads: -1
net.isr.dispatch: deferred