Performance penalty OpenWrt

We have a large number of TL-WR841N v 11.1 routers in our network that I would like to access remotely to run iperf3. I have installed OpenWRT 18.06.9. Total available memory is 23%,
When I use on a PC through the router with original firmware I get about 80 Mbps. Through the router running OpenWRT I only get 50 Mbps. Is this a performance penalty I should expect? Or could there be another reason?


Not exactly penalty, but proprietary software flow offloading on stock firmware performs better than the OpenWrt.


So I suppose there is not a memory issue. We have a number of TL-WR841N with version 13. These have apparently substantially more memory (8/64 vs 4/32). But that would not help, then, I suppose.
I would be looking for CPU power, I suppose.
Any suggestion what would be a reasonably priced OpenWRT router that will go up to 100 Mbps?

Ubiquiti EdgeRouters can take OpenWRT, apparently. I've used an EdgeRouter X, EdgeRouter Lite, and now an EdgeRouter 4, and have been happy with their performance.

However, I've stuck with the manufacturer firmware and haven't attempted to put OpenWRT on them. So I can vouch for the hardware and vendor firmware, but I can't vouch for the OpenWRT experience on that equipment.

If you don't need the wifi so much, the RPi has plenty of ram, cpu power, and 4 cores. You can get it at around 100€ including case, psu, gigabit switch, and a usb3 to ethernet adapter for independent wan link.

We are using a wifi network for our rural community. All of our wifi devices (about 160 or so) are Ubiquity. Ranging from nanostations to a few AirFiber radios.We are using a few EdgeSwitches, but for most of our switches we have move to Netonix and Mikrotik.
The clients are connected via a TP-LINK router. Primarily because of the cost effectivenes.
But I have not figured out a way to access the TP-LINK through ssh and/or install iperf on it. So far I've used Raspberry PIs. Which seems to work OK.

Which is why I've installed OpenWRT - but our clients will expect performance close to 100 Mbps. So I don't think this will work.
So what I'm looking for is an OpenWRT wifi router capable of passing 100 Mbps. For up to €30 or so...
(This may not be pssible at this time...)

MT7621A device like the Edgerouter X, Netgear R6250, Mi Router 4A Gigabit, etc. are good for 100 Mb ISP.

Running iperf3 directly on a low-end router will take a lot of the CPU just running iperf, so it is not a good way to test line speed except on a very slow line.

1 Like

The old adage "good, fast, cheap; pick any two" springs to mind... :wink:


Hi Trendy,
Just saw your suggestion for the PI. It does work fine.
But in addition to, the TP-LINK. Most people want wifi (that they expect top penetrate 60 cm walls...) and one or two LAN ports.
The PI's would be prohibitivley expensive for a large rollo out at this time.
Thanks for thinking along!

1 Like

Fast and cheap would do nicely...
As long as they last for a while. We have been replacing TP-LINKs as well.
But I get you point.

With all due respect, I think this may be a false economy. The rest of the equipment you've identified isn't cheap, and in the quantities you've mentioned you're looking at 4-5 figures in hardware costs. And yet the stumbling block is the cost of the router?

1 Like

There is another reason. Your router is both trying to pass the traffic and process it. Running a speedtest on a router gives false low results.

(Unless I misunderstood; and you meant that you plugged the PC directly into the ISP connection, then yes, sucky router CPUs seems like the culprit.)

1 Like

The Ubiquiti radios serve a wider area - the backbone consists of a few relatively expensive radios - AirFibers HD. But much of the other equipment and the CPE installed at the majority of our customers are using LiteBeam 5AC Gen2 and such at €50 ex tax.
We run our group with unpaid volunteers and take pride in offering good service at cost price. In fact, our 'service' ends at the network radio. We offer the TP-LINK as a free add-on.
Hope you understand.

1 Like

Gotcha. Makes sense. However, you're still up against the potential performance issue of inexpensive hardware.

One way to test the router performance reasonably accurately is to put one computer either side of the router and run iperf through the router, instead of hosting iperf on the router. The router's primary job is to move packets around, not to generate benchmark data.

@lleachii ,
I was running the speedtest on a PC connected to a LAN port of the. So the router would only be concerned with routing.
But the goal is to be able to test at the TP-LINK as and end-point (i.e. not routing). At that point it would be a simple Linux box connected to our network.
The link to test starts in fact at an Ubiquiti network radio. But, indeed, if I test there I get substantially lower figures.

1 Like

Sounds like you beat me to it! :smiley:

1 Like

The goal is the measure the speed of the link arriving at the radio at the customers premises.
On one end (30 km through the valleys and the over hills) we have a relatively powerful Dell server in the datacenter where we are connected to the fiber.
So I would like to test with a linux box connected through the Uniquiti radio to our datacenter.
So far I've done this with a PI between the network radio and the TP-LINK. Which works fine.
But I would like to combine this in a single unit.

Yes. Must say this is the fastest forum I've been on for a long time!

1 Like

I believe that final requirement is going to pose the greatest challenge, then. By trying to host the traffic generation on the device which is shunting the traffic around, you'll be diverting some resources away from the primary purpose of that device: to shunt traffic around. As a result, you're unlikely to see wire-speed performance in any tests.

I've just looked up the specs of the TL-WR841N and compared them to the cheapest EdgeRouter, the EdgeRouter X. The latter outstrips the former by a long way. In addition, the software on the latter (a derivative of Debian) happens to include both iperf and iperf3 already in the stock firmware.

If 30 Euros is your budget, you might get lucky and be able to pick up an EdgeRouter X cheaply on eBay or similar.

The main difference then is the wireless interface. The EdgeRouter does not have wireless where the TL-841N does.

I was assuming the system would do one thing at the time - route or run iperf. This will certainly not be valid always, but even it the system would be powerful enough with both tasks, the measurement would be upset, because the part of the download going to the client's direct usage is not counted. But that's ok. I normally measure once every 10 minutes 24/7.
But, as you say, finding a 2.4 GHz wifi router, capable of doing 100 Mbps with a few LAN ports, under OpenWRT is possibly a bridge too far. For now at least.

1 Like