At my house I have a T-Mobile ISP modem that gives me about 300Mbps down and 40 up when connected via the built-in WiFi or Ethernet. However, I do not use their modem directly as my main AP. Instead I have placed an Archer A7 v5 running OpenWRT behind it as a NAT and AP. Through this AP, I only see about 100Mbps down and 25 up.
I have been trying to find the bottleneck but I am not having much luck. When I run a speed test while monitoring "top" through SSH, I see the kernel software interrupt handlers taking most of the CPU, but the load average only reaches 30%, so I don't feel like the CPU is a bottleneck. The router has gigabit ports and all my cables are Cat5e, so I don't feel like they are the issue either. I have also tried swapping the cable between the router and the modem, but it had no effect. The same cable when plugged between the modem and a nearby computer achieves full speed.
What should I do to find the bottleneck? I appreciate any help.
A single-core 750 MHz mips 74Kc (QCA9563) CPU is not capable to do much more than 180-200 MBit/s, enabling software flow-offloading will extend that limit a bit (probably sufficiently for your current needs).
Just for a comparison, my slightly older tl-wdr4300 (560 MHz AR9344) will achieve ~175 MBit/s (without software flow-offloading), but that's already far beyond its abilities, as throughput is extremely jagged at that point (with latencies to the moon and back, so ~130-140 MBit/s are probably its sane maximum).
Well, you didn't mention if your'e running any SQM. I had/have C7's, which are very similar to the A7. Slight difference in main SoC chip, slightly slower CPU. And, had a 300/30mbit cable service.
I would see up to 270-280mbit peak on the 5ghz radio though, without any SQM. Running Cake, that's about where I ran out of gas... 110-120mbit, thru wifi. Maybe 130-140 thru ether alone.
And, if you get down to 0% idle time, yup, you're out of CPU. You can have a fairly low CPU%, and still be saturated, like it sounds like you're seeing with loads of SIRQ's. Top is confusing that way.
If you are running Cake or other SQM, I'd say that's about what you should be seeing, and your best bet is to run 0mbit on ingress/download (that turns off shaping) and just shape your 40mbit egress/upload. 40mbit is easy and the A7 won't break a sweat, and for me, that would yield 70% of the benefit of shaping both sides.
If you're not running any SQM, then there's some other problem that needs to be figured out.