Before starting to work in the evening, I'll summarize former results.
In the morning, I worked out how to balance the time between adding nodes and sending/resending requests. If I clear all requests after new node joins in, the transmissions will be affected. Thus, I preserve requests that are already dealt or being dealt and only the rest are cleared. In this way, short-cuts could be used by nodes. In order to guarantee the existence of short-cuts, further speaking, every node should find new nodes within L with high probability. Consequently, the factor L should be deliberately considered. The other problem I worked on is that how to decide whether the new node is an eager node. There should be some constraints, to find reliable neighbors to transmit across the network. Perhaps, it's also related to L.
In the afternoon, I consider more details of the scheme and justify whether it's feasible. Work proceeds slowly.
In the evening, I try to make use of the existence of new nodes. Try to design a method such that the dissemination time could be diminished. My intention is to create short-cut to cut off the depth of binary tree each node holds(cut to a constant level is the best). Or, balancing the requests, because, ideally, every node only serves constant number of nodes. Before being devoted to a thought, we should assume it's achievable and verify what it can bring? can it realize the desiring results? Or it's blind to think how to achieve the thought.
You have to fasten your CPU! Make good use of it.