posted Oct 23, 2016, 6:20 AM by Sami Lehtinen
updated Oct 23, 2016, 6:21 AM
- Bottleneck Bandwidth and RTT (BBR) congestion-based congestion control - Nothing new? Just send so much as the receiver can handle and try to estimate RTT and don't send too much. Yet these protocols might work well 'alone', but might now work well when there's someone using the bandwidth hog protocol I mentioned earlier. That's partially the problem. How do you know how much you can send. Some protocols try to minimize buffer bloat and others maximize it, when you then hit those shared buffers the ones minimizing the buffer bloat will get very small portion compared to those which are trying to maximize it. But otherwise BBR sounds really nice, when you can expect that everyone else is also using well behaving protocol like BBR. - Basic method for such probing is pretty much what they've used. Find out minimum RTT, then add more and more traffic, until the latency starts to grow. Then lower the traffic until latency is again at the stable minimum. - I've used that always when probing for 'free' bandwidth. But it doesn't mean that when the latency starts to grow that would be the point where there isn't more bandwidth to be robbed from other users by just adding stuff to the buffers. Even if latency grows, so can the bandwidth grow too, because programs using algorithm like BBR will give up and you can steal the bandwidth. Of course the same problem applies to any shared resource which should be used in 'sane moderation'. There always will be people who are going to abuse that. It would be nice to see charts how these different algorithms mingle together. - Interesting? Also check out CoDel - I did read more about the BBR topic on net, and almost everyone seems to be bringing same issues up which I did. That's nice. Also check out TCP congestion control.
- Microsoft is also bringing these features to Windows 10 / Server 2016 - TCP Fast Open (TFO), Initial Congestion Windows 10 (CW10), TCP Recent Acknowledgement (RACK), Tail Loss Probe (TLP) and TCP Low Extra Delay Background Transport (LEDBAT).
- I really like the Fido U2F standard, because now there are plenty of manufacturers on the market and many of those keys are considerable cheaper than the classic RSA SecurID Key Generators or YubiKeys. That is also the reason why I can't stop loving simple TOTP and HOTP.
- Windows 10 Anniversary Update contains lot of interesting stuff. Skype mess is horrible, but Bash is awesome! I'm going to love that for sure. I've so many times given bash commands in powershell. Haha.
- Anonymous Network, Tribler, quick analysis based on site & documentation. What are the factors I like and which I dislike and why. Based on P2P file sharing app, based on BitTorrent protocol (on some level). Streaming using magnet links. Content channels and reputation management. No centralized component. This is actually a good question always, how's boostrapping done if there's no central component? I've seen networks truly without central component. But it's very awkward for users, they have to give bootstrap information manually. Then the technical questions, create a dedicated Tor-like onion routing network. Does that even make sense? Tor is very inefficient way to deliver data. This is exactly where Freenet & GNUnet efficient packet caching steps in. Using three layers of relay proxies. That's horrible, first of all, traffic is still 'realtime' and therefore global advisory can monitor it and they're wasting a ton of bandwidth for nothing. Things get even worse when the seeder is also hidden, it's just like they would be using Tor. Two sets of of proxy pipes one for downloader and another for seeder. This means even more wasted bandwidth. So when you download 1 GB actually 7 GB of data gets transmitted over the Internet. In general, I don't like this kind of design, because it's so inefficient. With Tribler, after that 7 GB is transferred theres only one new seed. With GNUnet & Freenet, using efficient caching, there would be actually 7 new seeds after same amount of traffic. This makes the seeding & network availability go up much faster for new popular stuff. Also reminded my-self and discussed with a friends about Vuvuzela, Alpenhorn, Aqua, Pond, Dissent, Riffle and Riposte
- Something different? Yang Wang class tracking ship.