EAP655-Wall much slower downstream than upstream

This thread has been locked for further replies. You can start a new thread to share your ideas or ask questions.

EAP655-Wall much slower downstream than upstream

This thread has been locked for further replies. You can start a new thread to share your ideas or ask questions.
EAP655-Wall much slower downstream than upstream
EAP655-Wall much slower downstream than upstream
2023-07-01 02:29:51 - last edited 2023-07-17 14:12:28
Model: EAP655-Wall  
Hardware Version: V1
Firmware Version: 1.0.3 Build 20230329

To test my new EAP655-Wall, I've been measuring throughput using Librespeed and iperf3 running on my local network.

 

As a baseline, I confirmed a 1Gbps wired NIC connected to the downstream ports of the EAP655-Wall reaches line speed in both directions.

 

I've tried various wireless clients:

1. Laptop with Intel AX210 (PHY 2.4Gbps in TX+RX on 160MHz or 1.2Gbps on 80MHz)

2. iPhone 13

3. iPhone 11

 

with various settings, channels, and channel widths on 5GHz, and while upstream performance is satisfactory, sometimes reaching 900mbps, I am unable to exceed 450mbps downstream to any client, even at less than 2m distance and LOS to the AP. The performance is the same whether the AP is in standalone more or managed by a local Omada software controller. I upgraded to the 1.0.3 build 20230329 linked in another thread here to see if it would make a difference, but it's the same as on the 1.0.2 build that the unit shipped with. The 5GHz band is quiet here, with most neighbouring stations hovering around ch 36. I don't see anybody else above ch 44.

 

Strangely, when I go back to my Asus RT-AC68U (802.11ac 3x3) router, I can easily get to 500mbps in both directions which is reasonable given the 866mbps PHY.

 

I can clearly see that ax is working since my upstream to the EAP655-Wall is almost gigabit speed, but am at a loss to explain the lackluster downstream speed, which is inferior to the AC68U that I bought in 2014.

 

Is this normal/expected performance? Is there a beta firmware available that improves Tx power or throughput on the AP? That's basically where I'm at right now - the best signal I get off the EAP655 5GHz radio is -39 dBm, whereas the AC68U gets down to -35 dBm. Oddly enough, I'm reading -23 dBm from the 2.4GHz radio on the same EAP655.

 

Thanks in advance.

 

Note: I've opened TKID230715951 with support for this issue to see if they can investigate from their end.

  0      
  0      
#1
Options
1 Accepted Solution
Re:EAP655-Wall much slower downstream than upstream-Solution
2023-07-17 14:11:54 - last edited 2023-07-17 14:12:28

Just wanted to post an update. Support had me test with an iperf3 server connected directly to a downlink port on the EAP655-Wall and performance for the wireless client was much improved, approx 800mbps down and 1000mbps up at 2m distance LOS. This tipped me off that there may be something else wrong with my test topology.

 

I had not considered the impact of a 10G device sending traffic to a 1G access point and a client downstream who may or may not be able to communicate at 1G. When media changes like this, the intermediary device needs to buffer the traffic and "smooth" it out before sending it onwards. When these usually small buffers fill up, traffic is dropped and must be retried. This is a common problem in network engineering and from what I understand there are enterprise switches with deep buffers designed to compensate for this scenario.

 

If anyone else runs into this type of problem you could consider shaping the egress rate on the port your AP is plugged into to ~900mbps, which almost completely eliminated retries in my iperf3 testing and improved throughput (700mbps @ 3 meters from AP) where a 10G host is sending to a wifi client.

Recommended Solution
  2  
  2  
#5
Options
4 Reply
Re:EAP655-Wall much slower downstream than upstream
2023-07-03 03:25:11

  @slipperysloth 

 

slipperysloth wrote

To test my new EAP655-Wall, I've been measuring throughput using Librespeed and iperf3 running on my local network.

 

Can you share the result of iperf3 speed test?

Did you do that like this document mentioned?

Just striving to develop myself while helping others.
  0  
  0  
#2
Options
Re:EAP655-Wall much slower downstream than upstream
2023-07-10 13:02:33 - last edited 2023-07-10 13:17:52

@Virgo iperf running directly on the EAP655-Wall is not representative because it is bottlenecked by the limited CPU performance of the device.

 

INFO:

Client machine: Windows 11, Intel AX210, IP 192.168.10.5

AP: EAP655-Wall, firmware 1.1.90, IP 192.168.10.254

Server: FreeBSD 13, Mellanox ConnectX-3 10Gbps wired, IP 192.168.10.2

Distance from client to AP: 2 meters

Wireless connection info: PHY speed 2400/2400mbps, ch36, 160MHz

 

Testing against the AP:

 

PS C:\Users\x\Downloads\iperf-2.0.9-win64> .\iperf.exe -c 192.168.10.254 -P 2 -r
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size:  208 KByte (default)
------------------------------------------------------------
------------------------------------------------------------
Client connecting to 192.168.10.254, TCP port 5001
TCP window size:  208 KByte (default)
------------------------------------------------------------
[  5] local 192.168.10.5 port 59368 connected with 192.168.10.254 port 5001
[  4] local 192.168.10.5 port 59367 connected with 192.168.10.254 port 5001
[ ID] Interval       Transfer     Bandwidth
[  4]  0.0-10.0 sec   156 MBytes   131 Mbits/sec
[  5]  0.0-10.0 sec   193 MBytes   162 Mbits/sec
[SUM]  0.0-10.0 sec   349 MBytes   292 Mbits/sec
[  4] local 192.168.10.5 port 5001 connected with 192.168.10.254 port 39220
[  5] local 192.168.10.5 port 5001 connected with 192.168.10.254 port 39222
[  4]  0.0-11.0 sec   175 MBytes   133 Mbits/sec
[  5]  0.0-11.0 sec   172 MBytes   131 Mbits/sec
[SUM]  0.0-11.0 sec   347 MBytes   264 Mbits/sec

 

During the test, the CPU utilization reported by 'top' (sirq) exceeds 80%, meaning that the CPU chokes at around 300mbps of throughput.

 

Performing the same test against the server on my network:

 

C:\Users\x\Downloads\iperf-2.0.9-win64>iperf -c 192.168.10.2 -r -P 2
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size:  208 KByte (default)
------------------------------------------------------------
------------------------------------------------------------
Client connecting to 192.168.10.2, TCP port 5001
TCP window size:  208 KByte (default)
------------------------------------------------------------
[  5] local 192.168.10.5 port 59431 connected with 192.168.10.2 port 5001
[  4] local 192.168.10.5 port 59430 connected with 192.168.10.2 port 5001
[ ID] Interval       Transfer     Bandwidth
[  5]  0.0-10.0 sec   396 MBytes   332 Mbits/sec
[  4]  0.0-10.0 sec   364 MBytes   305 Mbits/sec
[SUM]  0.0-10.0 sec   759 MBytes   636 Mbits/sec
[  4] local 192.168.10.5 port 5001 connected with 192.168.10.2 port 47808
[  5] local 192.168.10.5 port 5001 connected with 192.168.10.2 port 58632
[  4]  0.0-10.0 sec   274 MBytes   230 Mbits/sec
[  5]  0.0-10.0 sec   172 MBytes   144 Mbits/sec
[SUM]  0.0-10.0 sec   446 MBytes   374 Mbits/sec

 

The upload speed (from my laptop to the server) is roughly twice the download speed when using a server on the other side of the AP for the test (client--AP--switch--server). With iperf3 (not shown) and enough parallelism I can get close to 1000mbps upload, but still no more than 450mbps download.

 

I am wondering if I got a defective unit - it's very tempting to exchange/return it if this is not normal. Thanks for your help.

  0  
  0  
#3
Options
Re:EAP655-Wall much slower downstream than upstream
2023-07-11 02:01:58

  @slipperysloth 

 

Uhm...the Iperf test result is a little slow, maybe it's a defective device, you can share the result if you get an answer from support.

Just striving to develop myself while helping others.
  1  
  1  
#4
Options
Re:EAP655-Wall much slower downstream than upstream-Solution
2023-07-17 14:11:54 - last edited 2023-07-17 14:12:28

Just wanted to post an update. Support had me test with an iperf3 server connected directly to a downlink port on the EAP655-Wall and performance for the wireless client was much improved, approx 800mbps down and 1000mbps up at 2m distance LOS. This tipped me off that there may be something else wrong with my test topology.

 

I had not considered the impact of a 10G device sending traffic to a 1G access point and a client downstream who may or may not be able to communicate at 1G. When media changes like this, the intermediary device needs to buffer the traffic and "smooth" it out before sending it onwards. When these usually small buffers fill up, traffic is dropped and must be retried. This is a common problem in network engineering and from what I understand there are enterprise switches with deep buffers designed to compensate for this scenario.

 

If anyone else runs into this type of problem you could consider shaping the egress rate on the port your AP is plugged into to ~900mbps, which almost completely eliminated retries in my iperf3 testing and improved throughput (700mbps @ 3 meters from AP) where a 10G host is sending to a wifi client.

Recommended Solution
  2  
  2  
#5
Options

Information

Helpful: 0

Views: 828

Replies: 4

Related Articles