I have experimented with different download and upload limits. Using the Waveform website I tested the results. I have achieved A+ in numerous download and upload tests. I have ATT Fiber with a 1Gbps upload and download service.
I noticed a setting of 930 on the download and 940 for the upload is the first achievable A+ score. However, the download rate cuts down to around 400 Mbps. Is this normal, for a massive cut in download speed where the download limit is set to 930ish?
When I leave the limit section blank and do several speed tests all is normal for about 980 down and 980 up. Curious if I missing something in settings?
A thing to note about CAKE is it doesn’t get a performance boost like a lot of other features on the Route10 by being offloaded from the CPU to the special offload engine. CAKE just doesn’t really mesh well with how the offload works as I recall. So the Route10 has to kinda brute force it with just the CPU.
That’s all to say that the speed reduction isn’t totally unexpected, although I recall some people getting much higher speeds that that when they first released CAKE support. Do you happen to have IDS/IPS on? I’ve noticed that combined with CAKE can really work the CPU in the Route10.
Another thing I’ve personally done is applying the limits on the port settings rather than the WAN settings. Limiting on the port itself yields some quirky performance as well from what I’ve seen.
Yes. Disabling hardware acceleration took the download speed back up to around 750 from the 380-400 average. Does this effect anything else in terms of network performance by having hardware acceleration off?
Great. Now, try tuning again, as a trade-off between lower latency, speed reduction and CPU increase during high bandwidth usage.
I believe it is all about CPU usage, and if it becomes a bottleneck due to network complexity, it might be worth backing off on the limit or limits or switch back to hardware acceleration.
Not bad at all! I would definitely be curious how it behaves with hardware acceleration back on but IDS/IPS off, just for the sake of science. As I recall, hardware acceleration on the WAN is supposed to be bypassed automatically when CAKE is enabled.
With HA turned back on it drops to 380-400 Download. Current limits set to 890 DL and 920 upload. Upload seems to stay at the same speed. However, when I turn HA back on it cuts the download in half.
Thanks! And that’s with IDS on or off as well? You mentioned it was on a few posts up so I was curious if turning it off specifically would make any kind of difference as well.
I did. Best settings for me are 900 down on limit setting and 950 upload limit. Keeps the latency low and also speeds are almost the same as previous tests. Any higher settings latency spikes and download speeds get lower. So the 10% down and 5% upload limits work well. Not bad at all. Acceleration is disabled. As soon as I enable it the download caps at 400ish.
I agree with, and can replicate your results. I have a bit higher unloaded latency due to my crappy usb-c to ethernet adapter.
Below is a screenshot with Hardware Acceleration disabled, and cake set to 10% lower than my 940 X 940 symetrical fiber. If I leave everything else the same and re-enable Hardware Acceleration, my speeds drop to just over 400Mbps up and down.
I’m seeing a similar issue but doesn’t fully correct if I turn acceleration off.
ISP speeds 1000/1000.
With accel on, IDS/IPS enabled, no wan dl/ul limit I get 933/940, Bufferbloat: C
With accel on, IDS/IPS enabled, 900/950 wan dl/ul limit I get 335/925, Bufferbloat: A
With accel off, IDS/IPS enabled, 900/950 wan dl/ul limit I get 741/930, Bufferbloat: A
Speed further improves with IDS disabled but I can’t get a test to complete (worked before now stalls)
I notice negligible system load on router during first test. The others go to 60-70%. I’m still losing ~25% of my bandwidth with accel off which is too much to me.
Is this just a hardware limitation? Seems to be based on a couple of threads?
Same question as previous in this thread. Is it possible to push the download limit further up while simultaneously considering CPU levels, latency and speed?
That’s with the limits set to 500Mbps set in each direction. System load registers as ~33% although I haven’t really dug deep into it. Hardware acceleration is enabled as well.
OpenSpeedTest: 580/560 (92/89% of nominal)
Speedtest: 560/570 (89/90 % of nominal)
System load is at 50-70 % down and 30-40 % up, for this type of test.
Cool! I’d like to add this to the list of quick speedtests to run as well, if anyone wants to try it out:
Made by the LibreQoS guys which I know was a project Dave Täht was putting some time into before he passed. Seems kinda similar to the Waveform test, at least in regards to utilizing the Cloudflare CDN for bandwidth purposes.
