5

In a WLAN iperf3 TCP throughput test, I have the following results:

With exactly the same setting, operation in 802.11ac (3x3) (DFS channel) (TCP Cubic)

30 nodes as server receiving iperf -> around 17Mbits avg per node.

30 nodes as iperf clients to a server -> around 4Mbits avg per node.

Is there any reason for this big difference between upstream and downstream?

They are peer-to-peer connections. The client/server or upstream/downstream concepts are application-layer concepts.

Is there something else in the TCP layer that is preventing the full link capacity from being used?

That's why I do not understand if the 802.11, IP and TCP layers are the "same", why there is such a big difference in the transmission rate.

Are there more collisions? Is that problem not solved with CSMA / CA?

Update:

What about ARF? This paper discusses the problem of uplink performance:

https://link.springer.com/chapter/10.1007/978-3-642-20798-3_28

Improve this question
11
  • "I know TCP is not a good idea over WiFi..." Where did you get such an idea? TCP neither know nor cares over which layer-1/2 protocols it runs. The problem with Wi-Fi is that it is a shared, full-duplex medium, and any throughput over Wi-Fi is going to be subject to the whims of the airwaves and sharing the medium. You will never get close to using the full, theoretical bandwidth of Wi-Fi the way you will on switched ethernet.
    – Ron Maupin ♦
    Jun 26 '17 at 21:37
  • I'm asking about differences between upstream and downstream in exactly the same medium. The goal is to test different WiFi solutions.
    – HenryGiraldo
    Jun 27 '17 at 6:56
  • What are you using as "client" ? If it's a smartphone, tablet or laptop, its radio power and antenna gain are limited; it will therefore send "less" than the AP which generally has more emission power, this might explain this difference if the signal is low. If you want same results, better use same devices as client and AP.
    – Benoit PHILIPPON
    Jun 27 '17 at 11:24
  • Using same devices as client and AP. 31 fully interchangeable nodes. All nodes are nearby and their MCS values are high.
    – HenryGiraldo
    Jun 27 '17 at 14:40
  • Understand that getting 50% of the theoretical or advertised bandwidth of your Wi-Fi equipment would be from nearly perfect conditions. If you have 31 devices, then you are going to get very low performance. Only one device can use Wi-Fi at a time, and the protocol requires that devices relinquish using the airwaves to give other devices a chance to use them. It is imposed sharing of a shared medium. Also. TCP is a bidirectional protocol of peers. You do not have client/server in TCP; that is an application layer concept.
    – Ron Maupin ♦
    Jun 27 '17 at 15:03
2

While passing downlink traffic (from the Access Point (AP)) to client-nodes, the AP schedules the packets from the packet-queues in the appropriate way and also does not face any contention on the medium.

On the uplink however (from the client-nodes to the AP), the client-nodes contend for the medium to send the packet. This creates a delay. This delay in turn reacts with the TCP timeout mechanism that triggers the TCP windowing mechanism to reduce the TCP window. This results in lower throughput.

If you really want to compare uplink and downlink throughputs, run the same test using UDP.

Improve this answer
2
  • I have tested with UDP and I also get really low performance in upstream. I think the problem is more related to ARF. "Auto-Rate Fallback (ARF) is a well-known mechanism in which the sender adapts its transmission rate in response to link noise using up/down thresholds. ARF has been criticized for not being able to distinguish MAC collisions from channel noise."
    – HenryGiraldo
    Jul 7 '17 at 8:10
  • @gpat Can you please elaborate on "and also does not face any contention on the medium." Why should the AP not be subject to the same shared media/half duplex restrictions as every other device?
    – Marc 'netztier' Luethi
    Sep 5 '17 at 11:06
0

In addition to the access/contention problem @gpat's pointed out, a wireless link is almost always asymmetric.

Each side is unique by location and very often by hardware ('cleanness' of the transmitted signal, reception sensitivity) and antennas. This creates a - more or less noticeable - difference in uplink vs download in terms of speed and error rate.

This can even get extreme when MIMO or MU-MIMO is used and the WAP can form a nice beam while the client can't and additionally suffers from parasitic deflections in its location.

Improve this answer
1
  • Exactly the same hardware, in direct line of sight and without interference. It's crazy, that's why it can't be understood.
    – HenryGiraldo
    Jul 18 '18 at 7:08

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.