Profiling node performance

With a group size of 3 and quorum size of 2, how many hops should be expected?

Currently routing splits every 1 MB chunk into 50 individual messages, and every group member sends a message for each chunk (either the chunk itself or its hash), which would be a factor of 150 for messages sent by a group, and a factor of 50 for messages sent by an individual node.
Then each chunk gets routed, so the number of hops would be the average route length (~ log2(network size)) times 150.
Every message gets an acknowledgement, so you’d have to roughly double that number again.

And then e.g. a Put requests goes from the client to the client manager and from there to the data manager, so that would be one node message and one group message, causing 400 times average route length hops, I think.

But there are also other messages sent by routing itself to learn about new nodes, establish connections etc., that have nothing to do with the data itself.

(I know, there’s a lot of room for obvious improvements here. But we have to make the network fully secure and resilient first, before we start implementing optimisations.)

7 Likes