วีดีโอ

What an incredible explanation. Thank you so much!

I don't think you should say that hashing the computer name, is the same as generating a random number, hashing it again will result in the same number, the hash isn't random, it just looks random.

Incredibly well explained. The algorithm reminds me of a theory spanwned some time ago, about the 5 grades of friendships. You are 5 acquaitances away from anyone on this planet

You did an amazing job of explaining the algorithm. What an ingenious system.

You are amazing! Thank you so much, this is really helpful!!!!

Damnnn

top-notch explanation, subd!

The stuff you mention at 11:30 about being non-Euclidean doesn't really make sense. Being symmetric is a requirement to be a metric and definitely doesn't make it non-Euclidean

this was soo coool !!!!

Now, what if you duplicate parts k-buckets over distributed systems, and suddenly you add in redundancy. If you can also add in "life expectancy probability" (for servers, i.e. for server up time, but also for server creation) into the calculation, then you can have a truly dynamic system, that can have servers pop up and "die" everywhere, without it mattering in the slightest for either up-time or service delivery times (outside of time units that are noticeable for human beings anyway). In other words, it's a super stable system that's very, very hard to break by any outside force; The kind of system that might even survive a nuclear attack, just like how the Internet was intended. ;)

I want to use this process to create a distributed and secure (as in non-censorable) messaging system. It's a direct reply to EU's murmuring about illegalizing all cryptography in the region, which is just an insane take on "security".

Ah so you just use XOR to count occurrences of "1s" and if there are too many 1s than some threshold, then the space is too close? Edit: When you find distances you basically find probability spaces at the same time. And you can check this with the probability space of other dictionary words, to make room for enough probable possibilities in every category? Incidentally, when you do polling, you'll immediately know when or if you're about to run out of addresses, so you can load-balance efficiently and start shovelling data over to other nodes/servers.

excellent explanation, thank you for making this video!

Honestly, the malware developer in me instantly goes; This would be an amazing way to check to see if a hashed password exists in my password list for brute forcing and cracking passwords.

thank you, this was a great explanation

This was such a clear explanation. Thank you.

I didn't understand discovery. Are you sure it shouldn't be the address of D as the hash key with routerD as a hash value? I'm not sure how it'd be used this way around but I'm probably just missing background.

congrats

I am shocked how simple and fascinating you made it! First time I see a video from you, but already your fan.

Really extremely insightful explanation <3 And the algorithm itself too..

Thanks for the video, I needed a refresher, and it worked grat :) There are some wrong details: - 2m55s: 4/8/16 bits are 16/256/65536 values - 5m16s: why have Dijkstra to start from 0? And why is the 'v6' stored twice? - 10m15s: the way you explain the xor distance, and the examples you chose, makes it look like it's |a-b| (see one of the comments), when in fact the xor measure gives a new definition of the distance which has nothing to do with the distance on the linear scale: 1000 and 0111 are 0001 away on the linear scale, but 0111 on an xor scale - 12m39s: 0101 xor 1101 is 1000 - 22m28s: the 'log' here is in fact 'ln' or 'ln2', so for 1e6 nodes, you need something like 20 hops. Still not too many, and still very impressive. And as you said, there are many details in the Kademlia protocol...

really cool algorithm! do you have any practical uses for it in your distributed systems work?

I stopped at 4:00 and now I'm currently working out the trivial details so I can come back to the next step of the video.

Dude you're my new favourite tech educator. Great energy, and very nice presentation

I don't really understand what the benefits of Kademlias K-Buckets are over Chord's Finger Tables. Seems like the same concept to me.

The dragon of the west? Sign me up

Just had my mind blown at 8:20 🤯

Bloom filters for the win! Thanks for sharing. This is great

I´m studing computer ingeneering and this is the best explanation i found about Kademlia. Great video, simple explanation, new subscriber. Thanks form Salamanca, Spain

Looks good, how does it compare to Holepunch?

Immediately subscribed because of the synth in the background.

Still no idea what this does lol

why xor is not working for 0001 xor 1000 = 9 but real distance is only 7

What is it though? Is it an OS? Web os? I’ll check the repo

Thanks. Best explanation of Kademlia ever

Feels like pied Piper

the lesson was very useful, but his green eyes.......💔💔💔

"No central server required" - that is a lie. I don't know how your stuff works, but you will need at least one server for signaling and at least one stun server and a turn server if you want to traverse any NAT.

wdym

can we integrate Iroh in Web ? building web apps using iroh and web frameworks ? like React, Svelte ?

I never even heard of a bloom filter before watching this series, but after I feel enlighted. Thank you for being so enthusiastic and teaching this in an easy way to understand

So basically reinvented p2p so traffic usage is offloaded to the users to reduce server cost?

Wait, I was just watching something relating to Bevy and saw the FishFolk Game as a bevy example of multiplayer. Funny to see it appear here as well. It take it Iroh is built with rust? Also based named. Uncle Iroh is based.

How's your middle-out compression?

How does it compare to Veilid?

Cool to see a tool for this, the devices of users aren't all equipped to store the entire state of their data so for some of my apps this would not be the final solution but having more interconnections that don't rely on a central server is always welcome. maybe this could be extended for federated software approaches as the activitypub protocol is still having a lot of issues

When I saw the channel name I immediatly thought that this was just another useless AI tool... But this is way better!

A little late to the party, but this is an awesome video and I have learned a lot from it. I have a small piece of feedback, though I’m not sure if implementing it would be helpful for your intended audience. I feel that your explanation of the triangle property leaves a little to be desired. At a surface level, the triangle property doesn’t have all that much to do with “triangulation” in the sense most people are familiar with, i.e. taking measurements from multiple places to narrow down a point’s location. **Way** down under the hood, the triangle property is important for triangulation, but only because it’s important for basically any process that involves geometry and metrics, and as far as I can tell the use of the word triangle in both of these concepts is more or less coincidental. Having not read the Kademlia paper, I have a guess as to how the triangle property may be relevant to the hopping protocol: basically a big hop plus a small hop lands not much farther away from the starting point than just the big hop. This assures us that we will never hop into the wrong distant k-bucket from our current node.

I probably need to watch one more time to understand the nuances, but I like the way you explain. Thank you

So what is it exactly?