Testing OpenWrt with the new ookla speedtest for testing latency under load

ookla has released a new app for testing latency simultaneously with load for ios and android.

https://www.ookla.com/articles/introducing-loaded-latency

How accurate it is, is unknown. I'd be very interested to know results with sqm on and off, and with fq_codel enabled chipsets vs not.

3 Likes

Hopefully they'll port that over to their desktop app for Windows.

1 Like

Amazing how many (one would think) unnecessary permission the Android app wants to have....

Speedtest takes up so much bandwidth loading ads that you don't know what the real speed is.

Not on the Desktop app.

you guys are tough. A test of this new app, with sqm on and off, on something, was all I'm hoping for. :confused:

1 Like

5GHz WiFi over my low end android phone is less than ideal for a speed test I expect. With SQM (CAKE) Speedtest result is 103/19 Mbps down/up with down/up ping of 100/20 ms. With SQM turned off, result is 107/23.3 Mbps and 99/56 ms on my phone. Distance to the Linksys EA8500 AP used for the testing was 10 feet of air.

My ISP service is a 400/20 DOCSIS 3.0 cable modem to a NanoPi R4S Gateway. Actual provisioned speed is ~450/23, but I've set ingress/egress for SQM at 400/20 to stay below cable service top speed during high use times of day. The EA8500 AP is connected to the NanoPi R4S with wired back haul through a GS308T managed switch.

From a linux thin client desktop with a 2x2 Intel AX200 card that is some distance away from the EA8500 AP with intervening walls, the desktop browser Speedtest result is 340/19 Mbps and 22ms ping (unloaded only) with SQM on; and 365/23 Mbps and 20ms ping (unloaded only) with SQM off. Obviously this is a better WiFi client than my Android phone :wink:

The Waveform buffer bloat test from my thin client (so again over WiFi) with SQM is 290/18 Mbps with mean ping of 33/38/29 ms (unloaded/down/up). As is typical, Waveform runs out of throughput before my ISP service or attainable WiFi speeds top out, and at least in my location, Waveform results are so variable from test to test and time of day that I rely on it for no more than a rough indication.

From a wired desktop, Speedtest results are consistently repeatable with and without SQM and obviously the CPUs in the NanoPi R4S do not slow anything down. But same issues with the Waveform test for me: repeatability is poor and reported down/up speeds are all over the place, except never approaching the consistently repeatable speeds indicated by Speedtest.

Interesting.

My Waveform results stay pretty consistent...but I test wired-only 99% of the time.

That's why I'd like to see Ookla port it to their Windows Desktop app.

The Ookla Speedtest is quite a bit more popular than the tests frequented by the denizens of this forum. So long as it is accurate enough to at least give a rough indicator, hopefully it will be a net "good thing" due to putting some positive visibility on buffer-bloat.

I test wired whenever I'm tuning settings. Full disclosure, I was too lazy to walk upstairs to a wired desktop for above post. Having now shamed myself into it...

Here are wired results with SQM on: Waveform 238/17 Mbps down/up and 28/30/29 ms unloaded/down/up; DSLReports 371/20 Mbps down/up 60/55/58 ms unloaded/down/up; and finally Speedtest 382/19 Mbps and 21 ms ping unloaded.

With SQM off, Speedtest reports 454/23.5 Mbps and 21ms ping unloaded. With SQM off Waveform is reporting anything from line rate (925 Mbps down) to 1.5 Gbps download today...so I need to retract that it never indicates as fast as Speedtest. However, I only have Gigabit ports and switches on the network, Cat 5E cable and I'm fairly confident my ISP hasn't upgraded me to 1.5 Gig LOL....at any rate, for whatever reasons, Waveform has been hit or miss for me.