| |||||||||||||
CeroWrt RC5 (beta) availableCeroWrt RC5 (beta) availablePosted Aug 22, 2011 0:19 UTC (Mon) by butlerm (subscriber, #13312)Parent article: CeroWrt RC5 (beta) available
Isn't it going to be a bit tricky to solve the most important bufferbloat problem there is (on the uplink side of a home / small business Internet connection) if you do not control the queue behind that uplink?
If you are not the bottleneck router, what can you do? Analyze all the connections going through your device trying to detect whether a queue is forming on the next one down the line, and throttle accordingly? Is that practical?
I would love to get one of these devices, but I am afraid they wouldn't do me much good, because I don't use wireless and the main bottleneck is the DSL router. I am hoping that this work will inspire the DSL modem manufacturers to clean up their act, or at least cut down the maximum queue size. I measured mine at 900 ms the other day.
(Log in to post comments)
CeroWrt RC5 (beta) available Posted Aug 22, 2011 9:11 UTC (Mon) by osma (subscriber, #6912) [Link] If you are not the bottleneck router, what can you do? Analyze all the connections going through your device trying to detect whether a queue is forming on the next one down the line, and throttle accordingly? Is that practical? If you know the uplink bandwidth and it's fairly constant, you can limit the outgoing data rate to the uplink to, say, 90% of the maximum bandwidth using standard QoS configuration. This way a queue won't build up, as the bottleneck is moved to the QoS-performing router. There's a tradeoff between bandwith and latency in choosing the optimum value, though. I don't know if any QoS-aware routers can do this automatically. It should be possible.
CeroWrt RC5 (beta) available Posted Aug 22, 2011 11:25 UTC (Mon) by mfleetwo (guest, #57754) [Link]
Gargoyle, derived from OpenWrt, has active congestion control which adapts to varying ISP upload bandwidth. There was also an LWN article about it.
http://www.gargoyle-router.com/
|
|||||||||||||