Matthew Toseland wrote: > The datastructure implementing this is limited to 10,000 keys > on each node (this will take up some RAM).
Hmm, so how much bandwidth does an attacker need to spend sending ULPRs for nonexistent keys before subscriptions for real keys start getting pushed out of the data structure, breaking up the subscription trees for those keys? The attacker needs to make 10,000 requests per hour to overflow a peer's data structure... less than 3 per second. Assuming ULPRs are 1000 bytes including overhead, an attacker with a 10 Mbps connection could affect 450 opennet peers. "Ultra-lightweight" could actually be a disadvantage here, because if the peers can easily handle that number of requests they won't throttle the attacker. Cheers, Michael
