My old setup was:
VSDL modem -> pfsense on mini J1900 Celeron (2 GHz) -> CISCO SG300 10MPP switch -> Rukus R310 wifi -> Laptop
Currnet setup
Fiber model -> pfsense on mini J1900 Celeron (2 GHz) -> CISCO SG300 10MPP switch -> Rukus R310 wifi -> Laptop
Today i got my 1GBit fiber installed (big deal for those like me living in rural areas) only to discover that my current network setup is not allowing me to benefit from it.
I was on VSDL copper wire before and was probably in the region of 50-60 MBit/s with my above current setup. Even when removing the wifi bottle and linking with Cat5 UTP wire directly to switch, I’m not getting major improvements.
When I got the fiber installed this morning I was disappointed when I saw only marginal gain running at 80 MBit/s (c. +30 MBit). So I decided to connect the laptop via LAN cable directly to modem. I got a starkling 900MBit/s. So, along my network I have bottlenecks.
THe first one I tested was my little pfsense machine. I installed the speedtext-cli command and was surprised to find that it was giving my around 300 MBit/s. So a lot better than my laptop on its usual wifi connection but still only 33% of what I get directly off the modem.
So my first question is how can it be that my little mini J1900 Celeron (2 GHz) with 4 GB RAM cannot handle this bandwith? Do I need an upgrade for my pfsense machine? I noticed that the peak CPU demand as speedtest-cli was running was in the 60% region, far from a saturated CPU and RAM only occupied for about 30%. If it is my little pfsense machine, how far do I have to go with finding the right little machine that can handle 1 GBit/s.
The next question is if I’m getting 300 MBit/s on the WAN connection of the pfSense machine, how is it that I only see a small percentage of this on my laptop? i.e. a drop from 300 MBit/s to 80 MBit/s? I guess I would have to test the switch to start and then move to the wifi access points …
You must log in or register to comment.
What can the network cards support?
Yes, e.g. rpi3b+ has gigabit ethernet, but it’s only 300Mbit, because it’s connected via usb2 internally. Something similar can be the culprit here as well.
Check what drop your get connecting the wifi modem directly to the router. There is usually a massive drop from wired to wifi.
The question is what you do with your pfsense. IDS/IPS are quite CPU hungry and Celerons are not really fast CPU’s.
peak CPU demand as speedtest-cli was running was in the 60% region, far from a saturated CPU and RAM only occupied for about 30%
It doesn’t look like he’s bound by CPU.
And he is currently at 1/3 of the potential speed and 3*60% = 180% CPU load for 1Gbits. So I wouldn’t even bother troubleshooting further when you already know the hardware will be an issue sooner or later.
That assumes that all of the 60% is for pushing packets, which is almost certainly not the case.
True. But since OP is using a benchmark anyways, I don‘t know how close to real world that is. If they are doing lots of filesharing, let‘s say with P2P networks, it could be way worse because of the number of connections. So I agree with you - I was just working with the info I had :)
Have you checked all the ethernet links are actually connected at 1G and not 100M?
FYI: Lots of the managed switches or the expensive wifi access points should be able to show the link status in their webinterfaces. It should be pretty easy to figure out if they’re running at 100M. (Sometimes also some LEDs light up in a different color.)
Yes, checked and are all on the 1000M (1G) link
The process is to go step-by-step. First direct connect to modem you have, bridged connection if possible, and test with multiple bandwidth measurements (speedtest, fast.com, downloading a big file from some university ftp…) and work your way downstream of the network. And on every step test multiple scenarios where it’s possible, preferably with multiple devices.
When I got a 1Gbit fiber connection few years back I got an Ubiquiti Edgerouter-X with PoE-options. On paper that should’ve been plenty for my network, but in theory with NAT, DNAT, firewall rules and things like that it capped on 6-700Mbps depending on what I used it for. With small packets and VPN it dropped even more. So now that thing acts as an glorified PoE switch and the main routing is handled with Mikrotik device, which on manufacturers tests should be able to push 7Gbps on optimal conditions. I only have 1/1Gbps, so there’s plenty of room, but with very specific loads that thing still is still pushed to the limit (mostly small packet size with other stuff on top of it) but it can manage the full duplex 1000Base-T. And on normal everyday use it’s running at 20% (or so) load, but I like the fact that it can manage even the more challenging scenarios.
Ok, starting to think I need a new little device for my pfSense. I was thinking of going OpenSense and buying one of their devices to support the project.
Regarding my switch, the ports where my Rukus APs are connected are showing 1000M on the interface. But I think a step by step testing is what is needed as suggested above.
They are expensive but I run a OPNsense DEC740 and have no issues with my Gigabit fiber, even without modem and the PPPoE overhead.
You can still try playing with hardware offload on/off and if you use PPPoE, it runs on a single core by default.
DEC740
Very nice but looks expensive. Do you think I could upload the pfSense configuration to it? I dread the pain of having to configure the whole thing from scratch.
I don’t think you can import pfSense configurations into OPNsense. I switched from a DIY pfSense box as well and redid the config.
You can look for a converter or install pfSense onto it though.
Run iperf internally to see if your bottleneck is switch/ap or fw. I set up a j1900 pfsense for my sisters family a while back to do qos (gamer bois in the house) amd it had no problem staying at 500mbps. No ids or other stuff.
Not built any opn/pf-sense in a while, but i always use intel server-nic’s. Used to have way better support than other stuff on bsd
Acronyms, initialisms, abbreviations, contractions, and other phrases which expand to something larger, that I’ve seen in this thread:
Fewer Letters More Letters AP WiFi Access Point NAT Network Address Translation PCIe Peripheral Component Interconnect Express PoE Power over Ethernet VPN Virtual Private Network
5 acronyms in this thread; the most compressed thread commented on today has 11 acronyms.
[Thread #861 for this sub, first seen 9th Jul 2024, 21:35] [FAQ] [Full list] [Contact] [Source code]
Do yuo have IDP/IPS turned on on pfSense? My OPNsense on my 1Gbps fibre will easily drop from an average of 900Mbps down to around 300Mbps-500Mbps, if I turn on IDS.
I dont’ have IDS/IPS installed on my pfsense box.
Something to look for besides bandwidth is actual packet routing throughput. It’s possible you enabled a feature (ex. Deep packet inspection) that is limiting how many packets can be routed per second given the speed of your hardware.
That Pentum is a budget CPU from just over 10 years ago. It has PCIe 2.0. Maybe the “gigabit” ethernet is connected to the CPU by a single 500Mbit PCIe lane.
500MB, not Mb. Order of magnitude difference there.
PCIe 2.0 is 500 MB/s per lane, it’s not going to limit the speed. That CPU certainly doesn’t have enough power to run something heavy like IDS at 1gbps though.
So my first question is how can it be that my little mini J1900 Celeron (2 GHz) with 4 GB RAM cannot handle this bandwith?
- check ethtool for link speed:
sudo ethtool enp2s0 | egrep 'Speed|Duplex'Your device name may be different fromenp2s0. useip linkto see all devices. if it’s not
Speed: 1000Mb/s Duplex: Fullthen that’s probably a bad sign.
- that is a 10 year old celeron processor. celeron were the budget (a.k.a. cheapest, slowest) class processor at the time. it’s quite likely that it cannot keep up.
- If you still think it’s not CPU directly, use iotop to see if you have I/O bottleneck.
sudo ethtool enp2s0 | egrep ‘Speed|Duplex’
Assuming you mean running these in the command prompt of pfSense? Tried but says “not found”. Same for “ip link”
pfSense is UNIX-based and those commands are generally included with Linux and probably Linux-specific.
Ah, didn’t realize pfSense is the OS, not something that runs on linux. My command examples won’t work for you.
- check ethtool for link speed:
I had to upgrade my pfsense hardware when I got fiber several years ago, which was in a similar situation as yours. The CPU just couldn’t handle the connection table.
J1900 has no hardware switch. Every packet goes through CPU, so even LAN to LAN uses processing power. Add pfsense to the mix and it’s probably choking.
how can it be that my little mini J1900 Celeron (2 GHz) with 4 GB RAM cannot handle this bandwith?
Because it’s ancient, and when it was new it was bottom-of-the-barrel.
I probably didn’t realise how CPU intensive the work of 1Gbit connection must be …
CPU and RAM are not the only limiting factors. Not only that but not everything runs multithreaded. Maybe some piece of the puzzle is not multithreaded and is using all it can from a single core (assuming that cpu is multi- core)
Depending on how much you value your time, you’re almost certainly better off getting a new machine to run pfsense.
Another piece of the puzzle is probably your WiFi router, as you normally won’t get speeds near 1Gbps over WiFi. In order to benefit maximally from it, you need to connect your devices (laptops, stationary PC, TV, etc.) with a cable to get the most of it.
You should also try to disable some pfSense plugins, like OpenVPN, zenArmor, etc. as they will severely limit your bandwidth throughput. But as others said, most likely you will also need to upgrade your hardware box, and you can migrate to OPNsense while at it.
For what it’s worth, since it sounds like you will be hardware shopping soon: I am using a 2.4GHz Intel Atom C2758 running pfSense and get 2Gb/s down and around 1.5Gb/s up through it. I am using an add-on Intel-based PCIe network adapter, so I’m not sure if that is helping with the CPU load. But it works well.
Any more specific recomendation of the machine you have running?
Maybe look into hardware offloading
What’s your ram speed?
