Flash crowd mitigation via an adaptive admission control based on application-level measurement

Flash crowd mitigation via an adaptive admission control based on application-level measurement

Chen, Xuan and Heidemann, John
USC/Information Sciences Institute

Xuan Chen and John Heidemann 2002. Flash crowd mitigation via an adaptive admission control based on application-level measurement. Technical Report ISI-TR-557. USC/Information Sciences Institute.

Abstract

We propose the network early warning system (NEWS) to protect servers and networks from flash crowds, which usually happen when too many requests are sent to a web site simultaneously. NEWS is an self-tuning admission control mechanism, which imposes application-level congestion control (AppCC) between requests and responses. NEWS detects flash crowds from changes in web response rate. Based on the application-level observations, NEWS adjusts the admitted request rate automatically and adaptively. Simulation results show that NEWS detects flash crowds within 10 minutes (about 2-3 detection intervals). By delaying 56% of requests, NEWS is able to reduce the packet drop rate for responses from 17% to 1%. The aggregated response rate for admitted requests is twice as fast with NEWS as compared to without. This performance is similar to the best possible rate limiter.

Reference

@techreport{Chen02a,
  author = {Chen, Xuan and Heidemann, John},
  title = {Flash crowd mitigation via an adaptive admission control based on application-level measurement},
  institution = {USC/Information Sciences Institute},
  year = {2002},
  sortdate = {2002-05-01},
  project = {ant, saman, conser},
  jsubject = {chronological},
  number = {ISI-TR-557},
  month = may,
  location = {johnh: pafile xxx},
  keywords = {NEWS, admission control, congestion control},
  otherurl = {http://www.isi.edu/%7exuanc/paper/news.pdf},
  url = {http://www.isi.edu/%7ejohnh/PAPERS/Chen02a.html},
  pdfurl = {http://www.isi.edu/%7ejohnh/PAPERS/Chen02a.pdf},
  myorganization = {USC/Information Sciences Institute}
}