This document discusses preventing distributed denial of service (DDoS) attacks that use web proxies. It proposes detecting abnormal request sequences from web proxies by analyzing the frequency of request sequences and comparing it to a web proxy's historical behavioral profile. When abnormal sequences are detected, a "soft-control" approach is used to reshape suspicious sequences rather than rejecting the entire sequence, to avoid impacting legitimate users. A hidden semi-Markov model is used to model the temporal and spatial behavior of web proxy traffic over time. This allows both fine-grained and coarse-grained detection of attacks at the server level, independently of traffic intensity or changing web content.