Do you have any ideas on useful routing statistics other than "number of hops" and "number of queries received by a node"?
I thought that maybe it would be useful to count the number of nodes that are also endnodes for a query, to check if the key distribution is really uniform.
Other thoughts?
Together to the avg Hops I have a bar-plot to show the occurrence of each hopCount.
E.g.
1 hop query - occurrence 3 times
2 hops query - occurrence 7 times
... etc
The plot should assume a shape of a "gaussian" where the peak is on the avgHopCount
Same thing for queries received: I count the occurrence of n. of queries received by each node
E.g.
0 queries received - by 90 nodes
1 query received - by 80 nodes
2 queries received - by 65 nodes
... etc
The (bar-)plot should assume the shape of a "hyperbola", at least it does for me for a sparse network (few nodes compared to the maximum number of identifiers available).
Does the latter look like "count the number of nodes that are also endnodes for a query" that you thought/suggested?
Or do you mean "1000 / 3400 nodes are endNodes for a query", indipently from how many queries received each of these 1000 nodes?
(Sorry in advance, I don't know how to quote!)
> Together to the avg Hops I have a bar-plot to show the occurrence of each hopCount.
This is actually a very good statistics, thanks for suggesting it!
> Does the latter look like "count the number of nodes that are also endnodes for a query" that you thought/suggested?
I understood that a node counts (generically) for a query if it is one of the hops, either a middle one or the end one. I was suggesting to count the number of nodes that are explicitly endnodes for a query.
Thank you, Andrea, for your suggestions. The first idea is to plot the distribution of the routing hops, and it would be very interesting to compare the experimental results with the analytical ones presented in the class. But, of course, this is not required in this mid term!
The number of queries received by a node is a good indication of the load balancing issue, so it would definitively be an interesting metrics to plot.
These are two different statistics. Counting the number of queries that each node receives is a good metric to evaluate the load balance/imbalance. As you said, counting the number of nodes that are endnodes for a query may be a check to verify that the key distribution is uniform.
Yes, I was suggesting to add the number of nodes that are endnodes as a statistics. :)
Should we include the endnodes in the hops' count? E.g. if a lookup goes 5 -> 21 -> 42 -> 3, should we count 3 or 4 hops?
I suggest 3, as the number of hops made by the query