Optimal Buffer Management

Krifa等人于SECON'2008上的文章：Optimal buffer management policies for delay tolerant networks。

关于全局信息m(T)和n(T)的获取，原文如下：

 

Our proposed solution is to find appropriate estimators for these utilities. We do this by designing and implementing a learning process that permits to a DTN node to gather knowledge about the global network state history by making in-band exchanges with other nodes. Each node maintains a list of encountered nodes and the state of each message carried by them as a function of time, which could be 0 if the message was in the node’s buffer at the specified time or 1 if the message was seen but deleted due to congestion as described in Figure 1. Note that each node maintains the time of the last list update and only sends the list if it has been updated since the last exchange. This way and after some time, all nodes will have the same global and accurate view about the network history. This history can be limited to some time duration if the network size is large.

 

基本思想是利用网络消息的历史传输信息来近似当前消息的相同参数，即：

Rather than looking for the current value ofmi(T) and ni(T) for a specific message i at an elapsed time T, we look at what happens, on average, for all messages after an elapsed time T. In other words, the mi(T) and ni(T) values for message i at elapsed time T are estimated using measurements of m and n for the same elapsed time T but measured for (and averaged over) all other older messages