I have three nodes that aren’t full, one of which I started yesterday. As with all my new nodes, I like to look at traffic figures to make sure that everything is working.
Comparing the new node to the other two non-full nodes (node C is new):
Node A: avg 119 KBps standard ingress
Node B: avg 120 KBps standard ingress
Node C: avg 30 KBps standard ingress
So the new node is receiving 25% standard ingress in comparison. This is still higher than the expected 5% for an unvetted node (not that I’m complaining) but comparing repair ingress traffic is way more interesting:
Node A: avg 36 KBps repair ingress
Node B: avg 37 KBps repair ingress
Node C: avg 67 KBps repair ingress
The new node is receiving nearly double the repair ingress traffic compared to the vetted nodes. Is there something in the repair process that prefers unvetted nodes as destinations for a repaired segment?
First of all measuring KBps throughput is a very bad measurement for usage of a node.
You should try to start counting the operations per action type. That way the piece size has no impact on your metrics
In terms of repair, i am pretty certain, that we do not differ between normal upload and repair (besides that fact that subnets of nodes already holding pieces of the to be repaired file are excluded).
Last time I checked the satellite’s code, it’s not exactly that a new node will receive 5% of traffic of a vetted node. Instead, at most 5% of all traffic is split between unvetted nodes, and at least 95% of all traffic is split between vetted nodes. What you observe is because there are many more vetted nodes (which compete with each other for the 95%) than unvetted nodes (which compete with each other for the 5%).
Some time ago I made a rough model for this phenomenon based on the source code at the time.
