We want you to try CAKE, to see if you're getting 1gbps then.
Thank you!
I have SEVERAL PS5 games downloading, I'm maxing my 1gbps and that's the CPU usage!!!!!! O_o
Then under those conditions do a videoconference with someone to see if realtime traffic is affected by the saturation of the line.
1 Like
I don't have anyone to call 
but I can stream a 4k movie, if that'll work?
My line is maxing out at 1gbps, PS5 downloads are getting downloaded very quick.
I'll play more tomorrow, going to sleep soon.
But for now, it seems SW/HW offloading really does work well. I haven't tried with SQM yet. (I did before but didn't have SW/HW on then)
that's what i keep saying since the begining of this thread, you don't need anything else. 
SQM is not recommended with Offloading. So you have to choose, you can't use both together.
Hardware flow offloading bypasses SQM.
3 Likes
Watching a movie, or youtube is not a valid test. All players prebuffer data in advance to avoid jitter, so it is not real-time and, in the end, it will get the bandwidth share it needs. But that won't test the worst case.
Another test you can do is play online games in GeForce Now, XBOX gamepass, stadia, etc. If you notice excessive lag between your input and the movement on screen then you can detect the effects of the bandwidth saturation.
In any case, a skype/Meets/Teams/Zoom/Whatsapp/Facetime video call is the fastest and easiest way.
With SW/HW offloading you get all the BW, but no prioritization. Latency under load should go beyond 100ms, and that is not acceptable for real-time communications or online gaming.
3 Likes
i agree, but for my use it works pretty well. My wife watching Live TV, and me playing online games on my computer, i don't see any problem at all. i'm not downloading and doing speedtests 24/24, installing SQM in my case will just cap me when i have something to download.
I'll stream a game with game pass ultimate then. see what goes wrong with that.
Good to know!
The real test for me, is the buffering on other devices. I'll see how I get on but for now, I'm impressed. 
1 Like
Actually you can also just download iputils-ping then run:
ping -i 0.1 8.8.8.8
Then saturate the connection by downloading a few iso files and verify there is no ping spike.
With significant bufferbloat this will render latency sensitive applications unstable.
I'm maxing my line atm with an XBOX download, I pinged my DNS which I'm using (8.8.4.4 & 8.8.8.8) and:
64 bytes from 8.8.4.4: icmp_seq=1 ttl=116 time=32.5 ms
64 bytes from 8.8.4.4: icmp_seq=2 ttl=116 time=31.7 ms
64 bytes from 8.8.4.4: icmp_seq=3 ttl=116 time=26.9 ms
Well I work with a 40-50ms baseline ping on my LTE connection even without any bufferbloat and Zoom/Teams works just fine, so if that is the extent of your bufferbloat then unless I am mistaken because my networking understanding is limited you should be fine there.
2 Likes
My bufferbloat with SQM has increased, have no idea why...
I've been playing with SW/HW offloading and it's great. Here's my CPU usage:
I'm maxing my 1gbps via a PS5 download:
The bufferbloat site is crap. I'm getting a ping of, when downloading the PS5 game:
64 bytes from 8.8.4.4: seq=0 ttl=117 time=25.121 ms
So, looks like I'm staying with this setup.
ping is not a good test, you can probably test hundreds of ping packets and all work fine, but for real-time videoconferencing or online gaming you need millions of error free packets to have a good experience during the video conversation or gameplay.
True but that still doesn't mean that you must use sqm, cake etc to get a good overall experience.
+1; nobody needs to use sqm/cake. This is a policy decision for each network to make individually.
Things that can improve with a properly configured sqm are:
a) latency under load increase, aka bufferbloat, but not all access links suffer from inaceptable bufferbloat to begin with
b) per-internal-IP fairness (using cake), this can help with fairly sharing an access link between multiple users, when bandwidth hungry applications are used (however sometimes such fairness is not actually desired, nor is flow-fairness).
c) competent ACK-filtering, on heavily assymmetric links the reverse ACK traffic for a download can get close to saturating the upload direction with ACK packets, that have quite some redundancy between them. This redundancy can be exploited to reduce the number of ACK packets without (noticeably) affecting TCP performance.
The main disadvantages are IMHO:
a) CPU cost, traffic shaping at higher rates is quite CPU hungry and can easily overwhelm an otherwise sufficiently powerful router
b) the strict fairness sqm offers is not the correct solution for all use-cases, sometimes targeted unfairness is desired and sqm offers not much in that regard out of the box.
None of these features is mandatory and/or universally desirable, all are tools in a network admin's toolkit that can help implementing the local policies.
Some people are happy without SQM, some are willing/happy to accept significant lower throughput when operating sqm on an under-powered router, none of these options is better or worse than the other IMHO.
Caveat: I am hardly unbiased in the question being a junior partner in sqm-scripts, but it is not that difficult to at least try to weigh the pros and cons somewhat objectively.
3 Likes
That ping test will already help detect sustained latency increases and will also give an estimate of the latency variance and tends to be a good indicator for bad conditions. But yes, an innocent looking ICMP train might coincide with sufficient short term jitter to cause issues in real-time latency sensitive applications, even though most of these already employ measures to be tolerant to some level of jitter.
Where the ping test fails completely is at assessing the intra-flow, self induced queueing of a capacity seeking data flow if a flow-queueing AQM like fq_codel or cake is used.
But such greedy flows are not your typical latency sensitive ones.... but e.g. cloud gaming which streams rendered video with adaptive quality on a narrow access link will fall into that category of bandwidth hungry and latency-increase intolerant traffic. Once the fair capacity share of a link exceeds such a flow's burst rate however, flow queueing will tend to do the right thing for such flows as well.
1 Like
Yep, I agree!
I agree with shdf now. you don't need SQM/codel to max your 1gbps with the RT3200.