Re: [Bloat] tsiox almost gets fixing bufferbloat right on reddit

[Rich Brown <richb.hanover@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1812 bytes --]

It seems that tsiox (on Reddit) is advocating that people invert the usual shotgun approach of guessing which QoS policies to implement, and instead use the following steps if "the network is slow":

- The first line of defense is minimizing queues in the bottleneck routers. Using fq_codel, pie, or other traffic shaping policy really makes a huge difference without spending time on configuration.

- After the traffic shaping is in place, measure the traffic responsiveness. If you can tell that the traffic shapers aren't quite doing the trick, then implement QoS, but only for the critical traffic that needs prioritization. 

- And of course, you may actually need more bandwidth. (The example in the Reddit thread of 100 video feeds at 1Mbps - you're going to *require* a 100 Mbps feed. No amount of queue management, prioritization, etc. will help.)

Questions:

- Is this a good policy, properly stated? 
- Is it something we should espouse on the Bloat/CeroWrt sites?

Rich


On Mar 7, 2015, at 8:26 AM, Dave Taht <dave.taht@gmail.com> wrote:

> I am nervous about applying sch_fq in this scenario rather than
> fq_codel but it is late - and honestly I have never tried what he
> suggests, can't remember my handle on reddit, and the guy otherwise
> gets it, in multiple postings on reddit elsewhere.
> 
> http://www.reddit.com/r/networking/comments/2xm5gs/qos_policy/
> 
> Can I ask that someone else monitor reddit, and chat with this guy, so
> I don't have to?
> 
> -- 
> Dave Täht
> Let's make wifi fast, less jittery and reliable again!
> 
> https://plus.google.com/u/0/107942175615993706558/posts/TVX3o84jjmb
> _______________________________________________
> Bloat mailing list
> Bloat@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/bloat


[-- Attachment #2: Message signed with OpenPGP using GPGMail --]
[-- Type: application/pgp-signature, Size: 496 bytes --]