2

I've been doing some testing lately on 802.11g and I'm very confused by the results. I'll go into the details below. First please understand I'm a software developer, but not a network engineer. I know very little about the lowest layers of how wifi or networking in general work. So part of this will be a learning exercise while trying to get an explanation for the behavior I'm seeing.

In my test I have 15 battery powered ARM devices running an 802.11g wifi chip. I run a script on my machine that sends an SSH command to each of the 15 devices, which executes rsync to pull down about 5GB of data.

The network is mostly Aruba equipment. I don't know a lot about it, but I know that the AP is their 205 model and for the test I had them configure it to 2.4GHz, because that's all the devices are capable of anyway.

When I run 1 device, it gets about 2MB/s. When I run all 15 devices, they all slow down to around 50-100KB/s per device, maybe more (it jumps up and down a lot, by a wide range, and is very inconsistent). Overall, total throughput across all 15 devices is 2-4MB/s. Even as I power down devices, the maximum of 2-4MB/s across all of them never changes. Each device ends up going faster as I turn off devices, and each device ends up going slower as I turn ON more devices. It's like there's some strange load-balancing happening somewhere.

My naive assumption was that since a single device got 1-2MB/s, that 15 devices would each get the same, for a total of 30MB/s or so. As I am told by others I work with, this isn't how it works. But they have no explanation as to why device throughput is being affected in real time when other devices get busy on the network.

Can anyone explain this weird load-balancing behavior? What it feels like to me is that the overall maximum throughput is the same for 1 device as it is for 15.

A subsequent test that I ran involved 15 completely different ARM devices with a wifi chip on them that supports both 5GHz and 2.4GHz, but since I restricted to 2.4GHz I believe they were running on 802.11n. At these speeds, single device runs between 5-10MB/s, but with 15 devices it sticks at a total of 8MB/s. Again, I'm seeing the same semantics here, but that overall, total limit is still there.

  • 2
    Wi-Fi is a shared medium, and only one device at a time can send. Wi-Fi forces the devices to take turns sending. Adding more devices will not increase the cumulative bandwidth. – Ron Maupin Nov 21 '19 at 16:23
  • So then what I'm observing is correct and expected behavior? – void.pointer Nov 21 '19 at 16:55
  • Yes. There are ways to increase aggregate throughout by adding more WiFi networks on different channels, but each channel can only be used successfully by one transmitter at a time within the radio range. Add too many active devices and aggregate throughput drops because of too many “transmission collisions”. One slow/obsolete or barely-in-range transmitter can also slow the shared network down for the other clients. – Darrell Root Nov 21 '19 at 21:01
  • You might want to look into multicast for a possibly more efficient data distribution across a shared medium. – Zac67 Nov 22 '19 at 7:28
3

Wifi is a shared medium. That means only one station can transmit at a time. When there is only one station transmitting, it shares the medium with the access point. The station (e.g., laptop) and the access point take turns sending data to each other -- they cannot both send data at the same time.

If there is more than one station, each one must wait until the medium is free before it can transmit. The more stations there are, the more likely one will have to wait (and longer, too) until it can transmit.

As an analogy, imagine a ferry carrying passengers across a river. If there is only one passenger wanting to cross the river, he or she can get right on the ferry (or at worst, wait for the ferry to return from the other side). But if many people want to ride, they have to wait for an empty spot on the ferry. The more people waiting, the longer it takes to get across.

The 802.11 protocol has mechanisms to give each waiting station an equal chance of gaining access to the medium.

| improve this answer | |
  • Late response, but what would explain why out of 15 devices, only about 4 are actively downloading but the other 11 sit at about 10KB/s? I've tried on different AP brands, the behavior is consistent. – void.pointer Dec 4 '19 at 1:59
  • Without knowing a lot of details, it's impossible to say. It could also be the application. – Ron Trunk Dec 4 '19 at 3:26
0

Wifi-6 will change that for us. This will allow all devices to get a chunk of the bandwidth depending on their needs, and there won't be a need for taking turns to talk to each device on the Wifi network, making it super efficient and super fast even when there are a lot of devices.

| improve this answer | |
  • 1
    Wi-fi 6 makes better use of (shared) air time but it still has serious limitations as to total bandwidth. If you really need bandwidth there's little alternative to a wired network - where real 1 Gbit/s per node/port (iwth dozens or even hundreds of ports) is nothing special, while that is next to impossible over Wi-fi for more than very few nodes. – Zac67 Sep 18 at 5:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.