CakeQOS Cake Status

[TonyK132]

I just ssh'ed into cake, and saw this status:
Sent 13260906455 bytes 12095560 pkt (dropped 11004, overlimits 3750932 requeues 0)

Does the overlimits mean what it implies, that I have my bandwidth settings wrong? I currently have it set for ~89% of my max ISP download speed, but because of variations, I suppose it could occasionally go lower.

 

[dave14305]

I’m pretty sure it just means that it limited the outgoing bandwidth to the configured rate. That’s the desired outcome to avoid bufferbloat.

 

[Find the Door]

From Merlin's notes on 386.1 (beta 3) change log


" - REMOVED: fq_codel support for Adaptive QoS. Due to a change
in how Trend Micro configures QoS, it is no longer
possible to intercept these to inject fq_codel."

I'd imagine this removal impacts cake? If so won't update from 384.19 does this change remove cake support?

Thanks

 

[brummygit]

Cake is still working well for me on the latest betas

 

[heysoundude]

From Merlin's notes on 386.1 (beta 3) change log


" - REMOVED: fq_codel support for Adaptive QoS. Due to a change
in how Trend Micro configures QoS, it is no longer
possible to intercept these to inject fq_codel."

I'd imagine this removal impacts cake? If so won't update from 384.19 does this change remove cake support?

Thanks

someone who's read the manpage, nice!
my interpretation is that while it relies on certain other techniques/schemes/code, it has rewritten/tailored them in specific/particular ways for the project, cake.
and in not engaging the TrendMicro stuff, we luckily bypass having any worries about breaking our things in upgrading

 

[tmcb82]

From Merlin's notes on 386.1 (beta 3) change log


" - REMOVED: fq_codel support for Adaptive QoS. Due to a change
in how Trend Micro configures QoS, it is no longer
possible to intercept these to inject fq_codel."

I'd imagine this removal impacts cake? If so won't update from 384.19 does this change remove cake support?

Thanks

Cake does not rely on the TrendMicro DPI engine to function, so it is not impacted.

 

[heysoundude]

Cake does not rely on the TrendMicro DPI engine to function, so it is not impacted.

This Is The Way

 

[Morris]

I just ssh'ed into cake, and saw this status:
Sent 13260906455 bytes 12095560 pkt (dropped 11004, overlimits 3750932 requeues 0)

Does the overlimits mean what it implies, that I have my bandwidth settings wrong? I currently have it set for ~89% of my max ISP download speed, but because of variations, I suppose it could occasionally go lower.

Drops are bad. They are what cake is designed to stop happening. You probably need to restore to default cake settings and reduce your bandwidth.

 

[TonyK132]

Drops are bad. They are what cake is designed to stop happening. You probably need to restore to default cake settings and reduce your bandwidth.

What is a good target for % Drops? After decreasing the download bandwidth, I'm now at about 0.8% of packets.

 

[L&LD]

0%

 

[Morris]

What is a good target for % Drops? After decreasing the download bandwidth, I'm now at about 0.8% of packets.

None. Try running a few long duration pings. An old AT&T target was 60 seconds error free. Can you achieve that multiple times? On a radio like like yours, there may be bursts of drops caused by interference and/or handoffs and then followed by long periods of perfection. You can't do anything about that. If the drops are due to overflowing the buffer, then you need to reduce the bandwidth more. Also run bufferbloat tests.

 

[TonyK132]

Thanks. I need to experiment with this. I just ran a few hundred pings and had 0% ping loss but cake was still reporting about 0.5% packets dropped. I also ran the DSLReports speedtest and got A+ bufferbloat when I had my Win10 PIA client off but with the client on, bufferbloat varied from A to C. But I just realized that my network was not completely idle when I did these tests. I still have a fair amount of continuous traffic happening on my 5G WiFi network through the router which might be impacting these results. Since this is real world for me, it is possible I may never get to 0% dropped with cake. But I'll keep playing with it.

 

[Morris]

Thanks. I need to experiment with this. I just ran a few hundred pings and had 0% ping loss but cake was still reporting about 0.5% packets dropped. I also ran the DSLReports speedtest and got A+ bufferbloat when I had my Win10 PIA client off but with the client on, bufferbloat varied from A to C. But I just realized that my network was not completely idle when I did these tests. I still have a fair amount of continuous traffic happening on my 5G WiFi network through the router which might be impacting these results. Since this is real world for me, it is possible I may never get to 0% dropped with cake. But I'll keep playing with it.

The link being busy should not cause drops. Cake should be able to slow responses and thus prevent drops which are recovered via timeouts. You can try cutting your bandwidth by 1/2 to 1/4. Cutting back on upload more drastically may not be harmful to your usage and solve the issues. As cell towers are shared, varying throughput can be expected and cake can only do it's thing when configured for bandwidth lower than the minimum you see. What you have right now might be a reasonable compromise. Look at user experience when streaming video and voice. If there OK, then you have done what is needed. Perfect grades on synthetic tests are great, yet a good user experience is what's needed.

 

[dave14305]

As cell towers are shared

What's the relevance of cell towers in this discussion? Is this because he mentioned his "5G WiFi network"? I took that to mean 5GHz WiFi, not 5G LTE.

 

[Morris]

What's the relevance of cell towers in this discussion? Is this because he mentioned his "5G WiFi network"? I took that to mean 5GHz WiFi, not 5G LTE.

I just looked back. I think you are correct Dave. I think I'm mixing this with someone that stated they are using a cell modem for backhaul. Even if it is WiFi, it's not as good a way to test as wired ethernet though some WiFi is excellent.

Tony, please clarify what you internet connection is.

 

[TonyK132]

All my testing was through a Ethernet connection from my Win10 PC to the AC86U. The router is ethernet connected to my Comcast Docsis modem. Most of the ethernet connections are 1G but I do have an occasional 10/100 connection. The PC ethernet connection is 1G. There is still a little network traffic happening on my 2.4G WiFi network and a lot of traffic on my 5G WiFi network. While most of that traffic is LAN only (cameras, using a 5G WiFi bridge to the router then Ethernet connected to a NAS), they will occasionally go to the internet. Neither WiFi network was involved in any of this testing. I do not use 5G LTE.

 

[Morris]

All my testing was through a Ethernet connection from my Win10 PC to the AC86U. The router is ethernet connected to my Comcast Docsis modem. Most of the ethernet connections are 1G but I do have an occasional 10/100 connection. The PC ethernet connection is 1G. There is still a little network traffic happening on my 2.4G WiFi network and a lot of traffic on my 5G WiFi network. While most of that traffic is LAN only (cameras, using a 5G WiFi bridge to the router then Ethernet connected to a NAS), they will occasionally go to the internet. Neither WiFi network was involved in any of this testing. I do not use 5G LTE.

What resolution are those cameras and what is the advertised speed of your network connection? How many cameras and what frame rate?

 

[TonyK132]

I'm not sure about the details of the cameras, but looking at Traffic Monitor in Merlin, I'm using about 32Mb/sec (4.0MB/sec) on the 5GHz network and about 5.6Mb/sec (0.7Mb/sec) on 2.4GHz, mainly going to the NAS.

 

[Morris]

I'm not sure about the details of the cameras, but looking at Traffic Monitor in Merlin, I'm using about 32Mb/sec (4.0MB/sec) on the 5GHz network and about 5.6Mb/sec (0.7Mb/sec) on 2.4GHz, mainly going to the NAS.

1080p is about 5-Mb per second. Not a factor for your wan link if going to your local NAS.

 

[TonyK132]

1080p is about 5-Mb per second. Not a factor for your wan link if going to your local NAS.

Yes. But all that LAN traffic might be impacting the routers performance when it is doing these tests.