Torture tests/usage for Asus/Merlin routers

[kvic]

I'm tempted to replace my all-in-one consumer router with one targeted for business i.e. more robust (to cope with harsh usage scenarios) and more capable/features (though most won't be needed in a home/SOHO environment).

To convince myself, I've to realise how good the low-end business routers are and how poor a home router can be. So I conducted tests to see under what workloads a home router will break. Apparently the tests I can do and dream up are very limited. Hence, this thread...to hear interesting stories from the fields regardless deployments in homes or in small business.

I think the use in small business is particularly interesting. I always wonder when a home router is deployed there, what kind of workloads it actually handles.

Here are my three simple torture tests

1. Linux PC connected through wire. One Linux client connected to 2.4GHz. Another Linux client connected to 5GHz. The 2.4G client send iperf streams (maxing out 2.4G throughput) to the wired PC. The 5G client pulls iperf streams (maxing out 5G throughput) from the wired PC.
   
   Result: the iperf streams can go on and on through AC56U without disruption. That's more than 40 hours continuously until I have to stop the test.
   
   
2. Linux server in the cloud with 200Mbit/s WAN. Home WAN 100Mbit/s. Home Linux PC connected to AC56U on 5GHz.
   
   Use scripts to initiate 12,000 iperf streams to the server in the cloud. This max out the 100Mbit/s WAN. Each stream will end up with a tiny amount, on average 8.3Kbit/s in steady state.
   
   Result: Active connections (as seen on the Sysinfo GUI) shoot up to near 12,000 and settle slightly above 10,000. I think that's more because of the PCs (actually VMs) at both ends aren't capable enough.
   
   For the ~10,000 active streams, they can go on and on again...for as long as I can bear with. This time around 20 hours before I've to stop the test.
   
   
3. Linux PC connected through wire. Bombard the little router with smallest packet and as many as possible. I shall say up front that my iptables table isn't that complicated.
   
   Result: AC56U can process ~67,000 packets per second without loss of a single one.

For all tests, I don't observe memory leakage. With Adaptive QoS turned on during test 1&2, I can even watch two live streams of video from Internet non disrupted. One stream at 1600Kbit/s on Apple TV. The other 800Kbit/s on a PC.

Also worth pointing out on a usual season that I don't tinker with the router, it can operate non-stop for many months. The longest I can recall is ~4 months. It was interrupted because I wanted (not that I need) a firmware upgrade.

I don't find an excuse to dump my all-in-one. Would love to hear others' stories.

 

[sfx2000]

You're only doing a single stream - use iperf to create multiple streams, and send/receive it from your remote source...

example - beat up the routing functionality..

5 streams...

iperf -c <other end> -p <port> -t 60 -P 5

as opposed to one stream...

iperf -s -p <port> -B <other end>

Throw in some packet corruption with netem... 2 to 5 percent...

 

[Nullity]

netperf is perhaps a better choice.

Thanks for sharing the tests!

 

[kvic]

Maybe try netperf or iperf3 on another time..

I was running 400 iperf2 on each side. Each process has 30 threads with one stream each using -P. Somehow I can't get -r nor -B to work reliably even between two processes with one stream.

 

[Nullity]

Playing with packet size might be useful to get a Packets Per Second measurement, for 64-byte & 1500-byte packets.

 

[kvic]

I used a quick and dirty ping flood test which is conveniently available. Ping an network interface on the router from a wired PC. To ensure it's routed and avoid built-in switch playing any smartness. I believe the tun (for ppl with OpenVPN server setup on the router) is a good candidate.

People can try with a script like this on your own router:

rm ping*.out
for p in {1..50}
do
    ping -q -s 1 -f 10.8.0.1 >> ping${p}.out &
done

Then aggregate the number of packets from all *.out e.g. in a spreadsheet. 50 is arbitrary here.

I can't remember how many processes of ping I need to reveal the maximum routing power of a 1.4GHz AC56U. The optimal number of processes shall be the one that produces maximum packet per second _at_ 0% loss of packets.