Config for many users

Hi all,
Bit of an unusual request for help. We are using GL-X3000 routers at over 500 sites.

At each site we have up to 600 (usually 1-200) Android tablets. When tablets are plugged in to charge at the end of the day (near the router), they:

  • try to connect to the router over WiFi
  • send small amounts of analytic data (under 1MB)
  • disconnect

We’ve configured openwrt on the router to a 2min DHCP lease time with 4000 addresses possible (and with a /28 subnet).

Not all devices will try to connect at the same time, but even now we see some clients just aren’t able to.

Any ideas on configuration options to enable all clients to successfully connect?

If you have more in-depth knowledge to consult and help us troubleshoot, please DM!

These don't appear to be supported by the official OpenWrt project. Therefore, the firmware you're currently using is almost certainly provided by GL-inet. That means you'll need to ask on their support channels/forums for help since they have heavily modified OpenWrt for their products.

2 Likes

Yes, and we've also posted there.

Still - if anyone here has any (more generic) configuration tips to help solve the issue, we'd really appreciate it.

Because the GL-inet firmware is so significantly different than the official OpenWrt, it requires specific and intimate knowledge of the vendor firmware rather than what is known about OpenWrt. Therefore, it is considered off-topic and out-of-scope here.

2 Likes

Thanks for your help, @psherman.

If any other readers have more in-depth knowledge to consult and help us troubleshoot, please do reach out.

This question is more about the classic hardware “i want a small router that can control the world but it must be cheap, very very cheap”

I think this has more to do about network math than OpenWrt or any other network firmware.

4000 leases doesn’t matter when you use a home router for 600wifi clients, or 200!? At pretty much the same time.

How many wifi clients can the actual wifi radio handle? On business class AP some manufacturers data sheets brag about 1000clients at the same time, with scrutiny of the commercial that is the control system max. The radio max is probably not even 200. I would say 50-100.

Then we have the problem with uplink speed spread out on these 200 or 600 clients…

But if you ask me, you have simply choked the network and the only meaningful solution you have is a business class network in obviously what is a business class environment.

To be clear, there aren't 4000 addresses available in a /28 subnet, there's about 14. Given that you expect 200+ devices to connect, (generically) I would assume this is your issue.

Feel free to donate:

1 Like

Thanks for all the help and advice so far, this is a great community!
A few further clarifications:

  • the subnet being used is actually /14
  • clients do not connect at the same time. At any given time there would be approx 20 connected simultaneously.
  • the cycle for a client is connect -> send 1MB data over the 4G wan connection -> disconnect

the issue seems to be that after a period of some time, new clients sometimes cannot connect. There don't seem to be any abnormally high CPU / RAM issues, leading me to believe it is a some kind of configuration issue, perhaps with DHCP.

wait ...
you are saying that you made one BIG L2 topology over 500 sites and put 4000 device in this L2 with one central /14 DHCP ?

2 Likes

No, this is per-site

Ok, it start to be confusing
but anyway ...
it is big commercial network, 500 sites, 4000 devices, data acquisition at end of the day ...
sorry to say, but ... seeking help for commercial products on free/volunteer forum ... :frowning:
good luck

1 Like

Configure the devices to retry a few times if upload of data fails.

Other ideas:
Log connection info on the router and analyze logs for more insight. Do the same for some clients, especially if you have units that are more problematic.

2 Likes

Thank you!
I'm going to try this solution and set dhcpleasemax to 4000. I understand the default is 150

For those wondering, this is not a commercial project, it is for monitoring schools using reading and maths apps in Malawi.

1 Like

The thing is that what you say makes no sense and is contradicting and you massively change the information after getting the answers!?

Why do you have 4000 active leases every 2minutes if you only have 20 connected clients at once?

And I doubt the wifi computers will let go of their connection once established and within range, they just reapplies every 60seconds if you give them 2minutes. The only way your idea will happen with dnsmasq is if the user turn off wifi on their device after uploaded the 1MB.

There simply isn’t a single operating system on any wifi device on this planet doing that kind of wifi connection you describe.

That's exactly what happens - devices enter airplane mode after transmitting data. We run a custom version of Android which triggers analytic data upload on plugging in to charge, then immediately disconnects from the WiFi network. So I guess there is one operating system that does this :slight_smile:

To be honest everyone sit in airplanes today with airplane mode on and Bluetooth headset on and if available on the plane wifi surfing on…

The only thing airplane mode kills today is the phone network.

And you have given these devices to kids/students…ok good luck with that wifi blocking.

1 Like

A valid observation!
In our context, children have no access to settings - this all happens programmatically.

Ok nice work, so why do have we this forum tread then, if your project work so good with 4000 leases for 600 clients but only 20 users?

I know a person irl that use to say like this: “don’t come and say how strong and heavy duty your tools are, throw them in to a kindergarten for a week instead and let us see how many seconds these heavy duty tools will survive, if a group of kids can’t break it during a week then it’s quality guaranteed really heavy duty shit.”

But since we have this forum tread to begin with obviously your heavy duty idea got a test fail from the kids proving ground and they have no laws and they don’t care for humanity when they are hired for doing standardized stress test…