CAKE regression?

[ebuckland81]

Checking in on a speedtest and seeing noticeably lower measured speeds than earlier, and wondering if there may be some regression going on, for CAKE applied, or if there is some intentional basic changes being made. I did try to turn off IDS/IPS without any difference. Also tried various settings for Acceleration, but it has to be set Disabled, otherwise throughout drops even more.

Nominal without CAKE is 940/940 Mbps on 1 Gbps ISP line. That is virtually what I get also on the WiFi client.

Now, at settings 825/875, I get

On Route10: 760/840 (top: 40-55 % idle)
On WiFi client: 490/820 (top: 65-80 % idle)

CPU is seemingly not the bottleneck.

[jmszuch]

I wonder if this could be related the change with UDP acceleration mentioned here?

Route10 Firmware 1.4l Released!

Ah, this is actually intentional. We’ve disabled UDP acceleration by default in this release because BitTorrents will waste too many flow slots on acceleration. They don’t have the bandwidth to justify acceleration, and chew through literally thousands of flow slots, even for just a single client. If you want to enable UDP acceleration again, you can run echo 4 > /cfg/alta_bits and reboot.

[ebuckland81]

I was contemplating on that too, but I thought Oookla speedtest ran TCP, so it shouldn’t be affected, unless I have misunderstood that part.

[jmszuch]

You might be right on that front! I think it was TCP the last time I tried to look that up, although I never went quite as far as running something like a packet capture to confirm.

I wonder if you’d see any different behavior with a test like fast.com or bufferbloat.libreqos.com (they use Cloudflare’s infrastructure I think).