 Greetings everyone from Switzerland and so this week during the launchpad program I've been focusing on measuring the actual IPFS network. This part is going to go very technical. I'm also going in Academy Implementation and KBuckets and Xoristance and all this stuff so if you feel lost, it's okay you can come back for the next presentation. So yeah, I strongly recommend to reading the Academy app paper to understand better the content that is coming to come. So I gathered and computed all of the data very quickly so it totally may be accurate and there may be some human error in it so please don't use the data or the graph as they are here I'm happy to provide more graph later on if you're interested. So the goals of measuring IPFS and more specifically the DHT and so the routing table of the DHT is to understand better how the routing works in practice in academia. So we know theoretically how it works because there was this great research paper but in practice I'm not aware of a deployment which is larger than IPFS for academia so that would be interesting to get to know more about. So we want to detect if some KBuckets that are missing some of the peers that should be included in this bucket which may result in degraded performance or we also want to measure the ratio of dead peers that are in the routing table. So if some of the peer are still in the routing table but are not reachable they are useless so we want to replace them by a new peers. And we also want to verify. So one of the property of Academy eyes that each note has to be aware of the 20 closest nodes. So we want to know if they actually know that the 20 closest node or if they are missing some of them and the goal in all of this is to understand how it works to propose improvement and eventually make IPFS faster. Some words about the measurement so I use the nebula crawler, which was built by Dennis an external collaborator, a very great tool if we want to do some measure or just explore the IPFS network. So we want to explore the data on April 27, around 10 ATC. And so, on the data so one measuring the wall, yeah, crawling the peers and their routing table, and the crawler was aware of 30,000 peers, and approximately half was online and half was unreachable at the time. I just wanted to show in the last few days to just make some graphs and I don't know how meaningful they are, but that's always nice to show in a, in a demo. So that's the feeling status of the K buckets. So, as you can see, it starts from year 220 so I did cut the first ones, because they are not useful as they're going to be empty and only the last buckets are going to be full. And then we can see so from bucket approximately 241 until 246. We have an exponential growth, and then it is kept to 20, which is a parameter of Calemria. And so we can see that for the smaller bucket, let's say starting from 226 to 241, we have a few exceptions that, so some of the neighbors know some peers that are really close in the shore distance. And when we go a step further, we can see then it's mixed so the buckets are not full but are getting full and on average it is quite stable as you can see the 50 percentile and so it's growing and so starting from the 248th bucket. So we have the, so the, yeah, even seven, yeah, everything basically is all of the buckets are full and the points you can see are outliers, but out of 30,000 peers, that's not a lot. That's the count for the then neighbors. So what we can see here so that's an average so a ratio which was amplified to fit the actual number of peers. So the first two peaks are just nice because it's where the nodes don't have a lot of contacts, a lot of peers in those buckets. And so if the peers of is offline, it's not going to be replaced and we're going to have a 100% of the neighbor, but what is more interesting is starting from 248th bucket. And we can see that it is lower and that's the data we actually want to measure. And so we want to understand, so that's the average of the neighbor for all of the peers. And we want to understand why those peers are not replaced. And it is good that this rate is low but we want to have it lower and to reach zero for maximum efficiency. Yeah, that's about what I have for now and I'm happy to hear more results when I have some.