fix: a bug in the dht size estimation, and add tests to prevent it in…

… the future

docs/dht_size_estimate/README.md

@@ -55,19 +55,15 @@ The final Dht size estimation is the average of `en_1 + en_2 + .. + en_n`
 
 ## Simulation
 
-Running this [simulation](./src/main.rs) for 20 million nodes and a after 4 lookups, we observe:
+Running this [simulation](./src/main.rs) for 20 million nodes and a after 12 lookups, we observe:
 
-- Mean estimate: 2,001,627 nodes
-- Standard deviation: 23%
-
-![distribution of estimated dht size after 4 lookups](./plot.png)
+- Mean estimate: 2,004,408 nodes 
+- Standard deviation: 10%
 
+Meaning that after 12 lookups, you can be confident you are not overestimating the Dht size by more than 10%,
+in fact you are most likely underestimating it slightly due to the limitation of real networks. 
 
-The relationship between the number of lookups and precision suggest that it doesn't take too many lookups 
-for the 95% confidence intervals to drop below +-20% from the real value. Meaning a very infrequent random
-FIND_NODE queries in the background, is all what an implementation needs to get a good-enough estimation of the size of the dht.
-
-![standrd deviation relative to number of lookups](./plot.png)
+![distribution of estimated dht size after 4 lookups](./plot.png)
 
 ## Limitations
 

docs/dht_size_estimate/src/main.rs

@@ -13,7 +13,7 @@ use statrs::statistics::*;
 
 const DEFAULT_DHT_SIZE: usize = 2_000_000;
 
-const DEFAULT_LOOKUPS: usize = 4;
+const DEFAULT_LOOKUPS: usize = 12;
 
 #[derive(Parser)]
 #[command(author, version, about, long_about = None)]
@@ -68,7 +68,6 @@ fn main() {
 
     for handle in handles {
         handle.join().expect("Thread panicked");
-        // println!("Worker joined.");
     }
 
     let estimates = estimates.lock().unwrap();
@@ -99,10 +98,10 @@ fn simulate(dht: &BTreeMap<Id, Node>, lookups: usize) -> usize {
 
             let mut closest_nodes = ClosestNodes::new(target);
 
-            for (_, node) in dht.range(target..).take(10) {
+            for (_, node) in dht.range(target..).take(200) {
                 closest_nodes.add(node.clone())
             }
-            for (_, node) in dht.range(..target).rev().take(10) {
+            for (_, node) in dht.range(..target).rev().take(200) {
                 closest_nodes.add(node.clone())
             }
 

src/rpc/closest_nodes.rs

@@ -2,7 +2,7 @@ use std::{convert::TryInto, vec::IntoIter};
 
 use crate::{common::MAX_BUCKET_SIZE_K, Id, Node};
 
-const CORRECTION_FACTOR: f64 = 1.26;
+const CORRECTION_FACTOR: f64 = 1.0544;
 
 #[derive(Debug, Clone)]
 /// Manage closest nodes found in a query.
@@ -77,7 +77,9 @@ impl ClosestNodes {
             },
         );
 
-        (CORRECTION_FACTOR * (sum / self.nodes.len().max(MAX_BUCKET_SIZE_K)) as f64) as usize
+        let count = MAX_BUCKET_SIZE_K.min(self.nodes.len());
+
+        (CORRECTION_FACTOR * (sum / count) as f64) as usize
     }
 }
 
@@ -163,10 +165,10 @@ mod tests {
 
                 let mut closest_nodes = ClosestNodes::new(target);
 
-                for (_, node) in nodes.range(target..).take(10) {
+                for (_, node) in nodes.range(target..).take(100) {
                     closest_nodes.add(node.clone())
                 }
-                for (_, node) in nodes.range(..target).rev().take(10) {
+                for (_, node) in nodes.range(..target).rev().take(100) {
                     closest_nodes.add(node.clone())
                 }