I made a comment at the mike at the IETF regarding CWV and data-limited periods, andSally asked me to send a brief note to the list. Briefly, the potential problem is that applications don't get scheduled with arbitrary granularity (10ms is not atypical), so they tend to generate data in a way which is not really continuous. When the RTT is small, it's pretty easy for CWV to kick in unintentionally - this is similar to what Colin's student Alvaro found with small RTTs. Anyway my point is that before we standardize any CWV behaviour for DCCP (at least behaviour that the app can't switch off), we do need to understand this interaction between app scheduling and DCCP over low-RTT links. - Mark
