I ran iperf3 from a hAP ac to Nano M5 on the same wired network segment, and got perfect results over and over again.
Then, I ran it from that Nano M5 to hAP ac, and got results that showed lots of retransmits.
I then repeated the test on another setup: A different hAP ac to a Rocket M5 on the same wired network segment. Perfect results.
Then, I ran it from that Rocket M5 to hAP ac, and got lots of retransmits, just like the first test above.
I would think that both directions would be the same?
Is there something up with the hAP's? I am running the latest beta firmware.
Both of these setups are using different hAP, different everything. Stock wall adapters.
Any ideas?
You are here
iperf3 results for A<>B vs. B<>A oddity
Sun, 11/03/2019 - 13:41
#1
iperf3 results for A<>B vs. B<>A oddity
Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer
Server: Rocket M5 Client: Mikrotik LHG-XL = lots of retransmits, and the tcp window never goes above about 50k
Server: Mikrotik LHG-XL Client: Rocket M5 = all OK
Clarify the 'wired' network is dtdlink cat5 cable with a configured 802.1q switch in the middle between the 2 devices? Or is this an RF linke between the devices?
Joe AE6XE
hAP #1 is wired through port 5 (dtd) to a power injector, to Nanostation M5.
I would expect the wired connections between hAP #1 and the NSM5 to have similar iperf3 in both directions, but the output of iperf3 shows about 500 tcp retransmits in the 10 second window, and a shortened tcp window, if the NSM5 is the 'server'.
If the hAP is the 'server' and the NSM5 is the client, iperf3 shows no tcp retransmits, and it will use the full tcp window in 10 seconds of iperf testing. I tried different cabling in the same setup, with same result. It's odd.
Is this result expected/normal?
The throughput is 'OK'. Iperf is showing 84-86Mb/s with the tcp retransmits, and about 88-93Mb/s when the test appears OK, so close to line speed of 100Mb/s.
Note however, achieving 84Mbps data 'throughput' with some retries is sufficient. The bottleneck is the RF side. The nodes on RF would be lucky to do half that in perfect conditions (on 10Mhz channel width). We're talking the application data thoughput, which is always less than the raw bit link rates. If inspecting any node on a roof/tower, and checking the Ethernet port, there's typically some small amount of retires, but this example is a bit higher retires, so something is going on.
Joe AE6XE
Thank you for your input.
The NSM5 is the XW revision. I have a Rocket M5 in a similar setup with the same results on the wired side.
If I take the iperf3 test one step further through rf link to a Rocket M5, I see iperf3 results with 95 tcp retransmits.
Then, if I make the hAP the server, and the Rocket M5 the client, I get no retransmits.
Rocket M5 ---<RF>-- NSM5--<ETH-->hAP
The Rocket M5--<RF>---NSM5 is working well in both directions.
I am going to continue to dig to see if I can find any hints. I will inspect cabling as you suggest.
Mikrotik LHG-XL <--> Mikrotik hAP on wired ethernet also does the same thing, no matter which is the server/client. Different wires, injectors, hAP than the example in post #4.
Really suspicious, does anyone else have any Mikrotik's that get the same results?
iperf3 must really be pusing the limits of the routerboard ethernet