Hi all,
I’m trying to find the cause and hopefully a solution to high latency under load. This all started due to a recent switch to full-time nomad life and mobile internet as my main connection. I picked up a MAX BR1 Mini and an external antenna and things were decent enough. I see 30/30 Mbps in good areas with unloaded pings somewhere in the range of 40-60 ms.
I was hoping to play some games over a streaming service (parsec, steam link), but found that I had a lot of stutter/lag. Troubleshooting that led me to bufferbloat, and running the tests from waveform and dslreports show my loaded download/upload latency typically around 1000 ms (but fluctuations probably from 500-2000ms). Tinkering with QoS and WAN buffer size seem to have no affect on this. Even a bandwidth limit of 0.5 Mbps keeps my loaded ping around 1000ms+. This is consistent with bandwidth limiting the steaming services to a 3 Mbps connection, and still having a terrible experience.
I also tested bufferbloat without the Mini just from my and my partner’s cell phones (AT&T, which the Mini also uses, and T-mobile), both of which return crazy high loaded pings too. So, it seems like the issues isn’t the Mini itself. Is this actually bufferbloat, and is this just how mobile networks are? In other words, is it just intrinsic to mobile carrier internet somehow?
If this is bufferbloat, is there any mitigation to all of this? The MAX BR1 Mini firmware, even on 8.2 RC1 doesn’t have the “mitigate bufferbloat” option. I assume a router running fq_codel or CAKE behind the Mini would help, but probably at the expense of bandwidth? But I’ve also read that SQM can be finicky on connections with fluctuating bandwidth (like mobile), is that right? Or would SQM not even matter in my case since the bufferbloat isn’t due to my hardware? (I guess it’s unclear to me if SQM can mitigate something happening on the ISP side, or if it just fixes things within your own network).
Any ideas or help would be appreciated! Really just trying to make sure I’m understanding things and evaluate what my options going forward are, a lot (pretty much all) of this is new to me.
I don’t see you saying it explicitly, did you enable Mitigate Buffferbloat on the support.cgi hidden page?
The problem is that fq_codel and cake all rely on fixed bandwdth systems. People are trying to adapt them to dynamic links with PIE type algorithms, but are always going to have feedback problems, because you have to cause a queuing issue, to then know that you are saturated and back off the limits. And that limit is often 1-2 network hops downstream from you, not directly connected.
Thanks for the input! I don’t think the Mini firmware (even on the betas or current RC) has the Mitigate Bufferbloat option. I’ve been through the support.cgi page on stable, 8.2 betas and RC1 and it doesn’t exist, unfortunately… but I could also be blind, so I’ll double check.
I have an unused netgear nighthawk r7000 running ddwrt that I’ll setup behind the Mini with fq_codel as my next thing to try. Even if there are complications with varying ul/dl speeds, I’ll at least learn something doing it.
You may want to try enabling “DSL/Cable Optimization” if it is available on the MAX BR1 Mini. This option was available before Mitigate Bufferbloat. I believe that it prioritizes acks going out on the upload/egress. On my Balance One, it’s under Network / QoS / Application / DSL/Cable Optimization.
FYI, Mitigate Bufferbloat works on upload currently. I believe getting it to work on downloads should be coming relatively soon. However, as Paul Mossip noted, “fq_codel and cake all rely on fixed bandwidth systems.”
At some point, it probably makes sense to use 5G where available, especially now that the higher frequency bands are being finally being deployed. 5G has better latency by design. And at least at this stage in the rollout of higher frequency 5G bands, likely has less contention.
