When I first installed preseem many years ago, I limited my house to 3 Mbps download for a few days as an experiment.
Netflix HD worked beautifully with no buffering. I then started a software download and preseem is designed to give equal bandwidth per flow and prioritize small flows like DNS, voip, web browsing, etc. So with one stream and a download going, video still worked but the adaptive bitrate reduced to SD quality with no pausing or drops. Then I went crazy and ran one Netflix, 3 YouTube streams, a software download, voip call, and monitored latency. All were sharing 3 Mbps total. All 4 video streams worked with no buffering. They all had SD quality as the available bandwidth per stream was under 1 Mbps. My loaded latency only increased like 4ms and everything worked fine. Voip didn't have any issues and websites still loaded decently fast. I remember using DSL or cable where if you maxed the connection, latency would spike to 800ms or higher. This is what bufferbloat is. If loaded latency increases hundreds of milliseconds, you will have a horrible experience. Using any modern rate limit protocol like fqcodel or cake will keep latency optimal whole maxing your connection. This helps things feel fast. If you have customers complaining of buffering with a 100 Mbps connection that isn't being maxed out, it could be one of many issues. In home wifi, old hardware like built in smart TV junk, congestion upstream of you, etc. Many of these complaints are solved when we tell someone to buy a Roku stick and STOP using the smart TV functionality. They use weak CPU's and wifi gear and rarely update the software. A good streaming device is key to a good experience. On Wed, Feb 26, 2025, 12:20 AM Dev <d...@logicalwebhost.com> wrote: > It seems traffic management pretty much has to happen, since it seems like > each streaming service wants something different, but basically some of > them say “hey, you have 50Mbps, I’ll use all of that then”, so they > optimize their service at the expense of the rest of the customers’ > traffic, but then there’s a big OS update on five laptops in the house, and > “now the internet is slow”. > > This means the ISP’s have to reverse engineer what the services expect and > come to some sort of middle-of-the-road approach for bufferbloat, bursting, > commit rates and the like inside the core routing, and probably keep > tweaking them over time. Very likely ISPs will consider that their secret > sauce and not share best practices. So we guess and test, guess and test. > > On Feb 25, 2025, at 7:55 PM, Ken Hohhof <khoh...@kwom.com> wrote: > > I’m beginning to think any sort of rate limiting is going to make one of > the hundreds of combinations of streaming services and devices unhappy and > customers will complain “Peacock is buffering on my Samsung smart TV” or > whatever. > > It seems like the world is going to 100+ Mbps best effort like 5G Home > Internet and Starlink, and streaming wants to ratchet up to 4K resolution > and burst at 100 Mbps. I can look at the Preseem metrics (they have lots > of them) and everything is perfect yet there will always be that one > customer who is complaining. It’s definitely not bufferbloat though. > > Maybe I should stop thinking I can please all of the people all of the > time. > > I do see the 5G providers saying they can limit video resolution to 1080p, > or allow 4K. Where they have two price tiers, that (and a mesh WiFi > extender) are often the differences between the two tiers. AFAIK, Preseem > does no DPI and can’t limit video resolution. With Net Neutrality dead for > at least 4 years, maybe I need to find a way to do that. > > *From:* AF <af-boun...@af.afmug.com> *On Behalf Of *Darin Steffl > *Sent:* Tuesday, February 25, 2025 8:30 PM > *To:* AnimalFarm Microwave Users Group <af@af.afmug.com> > *Subject:* Re: [AFMUG] bufferbloat at customer premise? > > Preseem is only one option. Free options do exist but they're more hands > on to integrate with your billing system. I wanted ease of use and the > pretty QoE data that shows us problem areas like congested ap's, wifi > issues, etc. It's worth the 50 cents per sub/month in my eyes. If we had > more subs then we'd explore an open source solution to save money. I'd say > for most wisp's under 3k subs, preseem or cambium QoE are worth the price. > > For us, it wasn't just about a good throttle solution like fq_codel, it > was the QoE data to help us resolve issues that customers don't report and > use it to confirm the fix. Preseem customer support is also very good. > > On Tue, Feb 25, 2025, 3:35 PM Dev <d...@logicalwebhost.com> wrote: > > Preseem might be more these days, but it seems like before they just ran > fq_codel or cake, which you can do for free in Linux, then charged per sub > so I have to rent my subs back from them. Lots of providers are trying to > get me to rent my subs back from them through cloud, ala Calix. > > > On Feb 25, 2025, at 11:42 AM, Darin Steffl <darin.ste...@mnwifi.com> > wrote: > > There's probably no need to do this at the customer level if you're > already doing it in the core. With Preseem or something similar, > bufferbloat is nonexistent. > > On Tue, Feb 25, 2025, 1:24 PM Dev <d...@logicalwebhost.com> wrote: > > Anyone using something like Mikrotik (or other vendor) bufferbloat queuing > at the customer router like fq_codel or cake? We’ve seen really good > improvement at the network core, but wanting to know if it helps to push > that out to the customer edge, and what algorithm/hardware does a good job. > -- > AF mailing list > AF@af.afmug.com > http://af.afmug.com/mailman/listinfo/af_af.afmug.com > > -- > AF mailing list > AF@af.afmug.com > http://af.afmug.com/mailman/listinfo/af_af.afmug.com > > > -- > AF mailing list > AF@af.afmug.com > http://af.afmug.com/mailman/listinfo/af_af.afmug.com > > -- > AF mailing list > AF@af.afmug.com > http://af.afmug.com/mailman/listinfo/af_af.afmug.com > > > -- > AF mailing list > AF@af.afmug.com > http://af.afmug.com/mailman/listinfo/af_af.afmug.com >
-- AF mailing list AF@af.afmug.com http://af.afmug.com/mailman/listinfo/af_af.afmug.com
