Just bought an cisco 3850 with mgig and upoe, because i want more throughput on wired and wireless.
Got the 3504 and 9120 connected to mgig on the 3850, set the speed to 5000 on the 3504 via the 3850, also set the speed to 2500 on the 9120 via the 3850.
3850 was set to system mtu 9020.
9120 was set to 160mhz.
But i dont see any inprovement at all on wireless.
Did also just buy an 2.5g mgig NIC for my pc, but i only went from 1g to 1.5g.
All test were done using iperf3 running on my server connected with 10g to 3850.
I'm uploading configs for 3850 and 3504.
i'm also uploading pictures with my results from iperf.
Picture 1-2: iphone 12 mini iperf3 download and upload
Picture 3: Dell latitude e7270 with intel ax200
Picture 4: Fujitsu p420 with Asus PCE-C2500
Picture 5: HP z620 with intel x520 10gig (not mgig)
Best greetings from Christian Josiassen.
WiFi perspective, there is a huge difference in Data Rate vs Throughput. Also, it all depends on client capability (the below page may help you to understand different clients' capability)
If you want to test the best case scenario, there should be only 1 client (iPhone in your example) on your AP & it should get higher RSSI/SNR to achieve the MCS11 data rate (1.2Gbps - assuming 80MHz, 2 spatial streams with OFDM). Even such data rate, throughput value is normally 40-60% of data rate, as there is a lot of overhead in a wireless cell. Therefore getting 506Mbps (upload direction) indicate you are getting close to the best possible values.
Download speeds are not exactly what you would like to be, I would do further testing by checking RSSI/SNR "show client detail <mac_addr>" CLI output tell what AP thinks client RSSI. If you can get those details from Apple client itself, that is what matters and that determines the MCS rates the client go with.
*** Pls rate all useful responses ***
You haven't answered Leo's questions about software and driver versions ...
For the wired connection try an updated driver directly from the chip maker:
Multi-Gig cards are very driver sensitive.
For iperf, also try it with TCP traffic, UDP is very hard on the CPU in some cases (CPU load would rise on one core up to 100%), while TCP allows many newer features lowering the CPU clock. iperf is usually single threaded, which is why only one CPU core gets loaded. I think there was an option for parallel streams and threads, to even out CPU load problems.