5 ms·
> customers paying for a guaranteed percentage of high-QoS > bandwidth during peak network load This and all similar ideas are very dangerous. Compare pure co
by dgregd 14y ago
> customers paying for a guaranteed percentage of high-QoS
> bandwidth during peak network load
This and all similar ideas are very dangerous.
Compare pure commercial networks (mobile SMSes, ISDN) with Internet which has academic background.
What is the right direction for Internet? Why we talk about paid QoS packets ideas and not about Google Fiber?
- nitrogen 14y agoI would definitely have more bandwidth rather than traffic shaping, but I also want my telemusic, teleconferencing, and gaming packets to arrive at their destinations ten milliseconds ago, and not get dropped the same way my bulk HTTP transfers should during heavy load. Even with plenty of bandwidth, there's still the buffer bloat issue. If you decrease the buffer sizes and use random packet dropping, your VoIP and gaming packets could get too many dropped. If you leave the buffers as they are, you could have very high jitter. So, either we need a way of saying we want high-priority packets not to get dropped when the buffers are full, or we need a way of saying we want high-priority packets to bypass bloated buffers.
- wmf 14y agoI think AQM (e.g. CoDel) + ECN is the solution to that. TCP will occupy very little buffer space, leaving plenty for real-time traffic.