"Five (5) 10/100/1000 (1 WAN and 4 LAN) Gigabit Ethernet ports with auto-sensing technology"
I've tried r48296, r51362 and a few others with pretty consistent results. So far, all I get when doing iperf3 between my PC and this router via the LAN ports is ranging approximately 100Mb/s. This is pretty consistent. Based on the dd-wrt Setup >> Switch Config page on the router, the port is connected at 1000Mb/s.
I have another gigabit switch (which is in the middle), shoes both my pc and the router connected to it at 1000Mbps.
Running iperf3 from my PC to another host (also connected to that switch) gives me around 995Mb/s speeds.
When this PC is connected directly to the router (bypassing my gigabit switch) also gives me ~ 100Mb/s speeds.
I certainly don't expect gigabit speeds on this router, but the consistency of 100Mb/s out of seems to suggest that it's somehow being restricted.
The CPU utilization (using top) on the router is at ~ 50-60% when iperf3 is sending/receiving data.
I've done searching, to no avail; there's been talk about CPU utilization, replacing cables and whatnot, but that doesn't seem applicable in this case. The factor in this case seems to be the router and dd-wrt.
Anyone know what could be limiting the speeds on this router?
I’m going to go back to stock software to try, but I suspect the speeds won’t be restricted as they were not before dd-wrt.
Posted: Thu Jan 19, 2023 3:52 Post subject: Solved?!?
Ok ... interesting observation.
When I have two hosts connected to the router's switch (the LAN ports), one running iperf3 (server), and the other iperf3 (client), I can get speeds of around 944Mb/s with dd-wrt running.
However, when iperf3 (server) is running on the router itself, then the speed is limited to 100Mb/s.
.. and yet, when the PC is running iperf3 (server), and the iperf3 client is the router itself, the speed is limited to 66Mb/s.
Anyone know if this is expected / seen across other routers too? I'm quite surprised tbh.
Posted: Thu Jan 19, 2023 5:38 Post subject: Thanks!
Thanks so much for confirming!
I thought that’d be the case, given the direction difference in throughput.
I just didn’t know enough about its inner workings. I imagine that the bulk of data switching isn’t really handled by the CPU on the LAN switch, which is why it’s not impacted when iperf is routed between the two external hosts connected on the ports. I’ve also read that with this switch, NAT severely impacts throughput due to cpu utilization, so I’m guessing there’s a lot more processing going on during the routing between WAN and LAN packet transfers.
This point on its specs is kind of telling:
"Allows data packet inspection to be offloaded from a product’s Central Processing Unit (CPU), thereby freeing up CPU bandwidth to provide superior performance."
So it seems to me that the intuition was at least remotely accurate.
Much of the packet switching is done outside of the CPU. Also, the main CPU is only running at 600MHz. There's zero chance you'd get gigabit speeds coming out of there, when you consider it has to generate the data to begin with.
To be honest, I'm surprised I got what I got out of it.
Anyways, no point in beating a dead horse. Thanks again for all of your help!
When I tested various routers a couple of years back, none of the custom ROMs operates the WAN interface at full duplex, only half. Stock ROM always worked at full duplex. I tested using iperf2's bidirectional option.
If you're connecting to a full duplex fiber connection, you're better off with stock ROM if you're expecting full duplex performance. If you're on cable modem or custom ROM features are more important, then half duplex it is.
