dtaht

14 ms·

181.
by dtaht 4y ago
But we wuz there, first: http://the-edge.blogspot.com/2003/06/wireless-connection.htm... I wish more thought about making wifi better. It could be so much more awesome now if we just sank more thought and resource
182.
by dtaht 4y ago
I have tried to fix the chipsets most used in openwrt. Presently the ath9k is best supported, the mt76 is pretty good, the mt79 coming along, and the ath10k, ok. https://www.youtube.com/watch?v=Rb-UnHDw02o&t=1560&the
183.
by dtaht 4y ago
I wish more understood wifi: https://www.youtube.com/watch?v=Rb-UnHDw02o&t=1560&themeRefr... enough to help fix it.
184.
by dtaht 4y ago
Not a lot of people know that linux´s original network stack more or less came from phil karn giving permission to copy KA9Q.
185.
by dtaht 4y ago
Thx for the heads-up greg. Nice to see that old pre-blog entry again. I wrote a 3 and 10 year retrospective of the project here: http://the-edge.blogspot.com/2010/10/who-invented-embedded-l... I never found out if
186.
by dtaht 4y ago
We used to use old DEC alphas as firewalls figuring the arch was so obscure as to not worry about binary attacks.
187.
Flaws and features in the flent network measurement tool(blog.cerowrt.org)
1 points by dtaht 4y ago|0 comments
188.
by dtaht 4y ago
My principal reason for proposing that 127 be reduced to a /16 instead of a /8 was to help unwind the kubernetes hairball. I have no idea to what extent 0/8 is being used today. I do see quite a bit of 240/4 in private n
189.
New features and flaws of speedtest.net and Cloudflare(blog.cerowrt.org)
4 points by dtaht 4y ago|0 comments
190.
by dtaht 4y ago
There have been multiple attempts at making 240/4 globally useful. It pretty much just works on most gear today (except windows), but the resistance by the IPv6 crowd to opening up 1/16th of the ipv4 internet for general use, asto
191.
20+Gbit Net Neutral QoE for ISPs Using Cake(libreqos.io)
9 points by dtaht 4y ago|0 comments
192.
by dtaht 4y ago
It's not rocket science for starlink to make their bandwidth better. It's IETF rfc7567, rfc8289, rfc8290, & rfc8033, published many years ago.
193.
PCMag – Starlink Speeds, Congestion and Bufferbloat Wo(old.reddit.com)
2 points by dtaht 4y ago|2 comments
194.
by dtaht 4y ago
I really wish y'all had a grip on and were measuring the bufferbloat you are creating with TCP_INFO. hundreds of flows in slow start are really tough on the network even with fq_codel or cake, and ledbat...
195.
Teledesic in Leo in 1996(web.archive.org)
1 points by dtaht 4y ago|0 comments
196.
SamKnows – US DSL has 853 ms of latency under load(samknows.com)
2 points by dtaht 4y ago|0 comments
197.
Amazon, Verizon found using IPv4 240/4 addresses(labs.ripe.net)
193 points by dtaht 4y ago|143 comments
198.
Turris Omnia home router project fails to spin up and out(linkedin.com)
2 points by dtaht 4y ago|0 comments
199.
by dtaht 4y ago
It's a shame "speed" is even a benchmark. Over here is a NTIA filing, with some current bufferbloat related working latency data for starlink, dsl, 5G, and fiber in the appendix. https://docs.google.com/docume
200.
by dtaht 5y ago
I have long worried that net neutrality rules mis-applied would prevent the default availability of quality fair queuing and AQM solutions along the edge of the network. https://blog.cerowrt.org/post/net_neutrality_cust
201.
by dtaht 5y ago
The director of the NTIA $70B broadband expansion program is taking questions wednesday in a fireside chat. https://discuss.broadband.money/c/broadband-grant-events/ala...
202.
by dtaht 5y ago
I long ago made bcp38 and fq_codel available in openwrt. It would be great if cloudflare told more customers what better home routers they could use as a base. We also solved the bcp38-like problem ipv6 had by using source specific routing
203.
by dtaht 5y ago
Very researchy. fq_codel and pie are deploying now.
204.
by dtaht 5y ago
keep hoping more deploy rfc8290 by default.
205.
by dtaht 5y ago
I am collecting suggestions for next generation edge routers here: https://forum.openwrt.org/t/cerowrt-ii-would-anyone-care/110...
206.
by dtaht 5y ago
google wifi also has the same option, using fq_codel, almost as equally hidden. IMHO having it on by default would improve both products out of the box experience. UBNT now does it by default, at least. Happy to have seen both cake and fq_c
207.
by dtaht 5y ago
I wish eero's sqm was more configurable and adequate statistics were available. sch_cake, as used on the eero 5, and in the modern openwrt SQM world, is definitely proving out better for many workloads.
208.
by dtaht 5y ago
I'm always curious how many users use the smart queue management feature (being one of the authors of fq_codel), or are aware that's also the default on the wifi?
209.
by dtaht 5y ago
fq_codel is the default on linux, osx, and ios now. It "just works" at line rates. It's also part of the linux wifi stack for a half dozen chipsets. It's not visible via qdisc as it's at the wifi layer: https:&#x2F
210.
SpaceX Starlink deployment in Ukraine – one week later(circleid.com)
3 points by dtaht 5y ago|0 comments

More ›