Detecting ICMP Rate Limiting in the Internet

Detecting ICMP Rate Limiting in the Internet

Guo, Hang and Heidemann, John

Hang Guo and John Heidemann 2017. Detecting ICMP Rate Limiting in the Internet. Technical Report ISI-TR-717. USC/Information Sciences Institute.

Reference

@techreport{Guo17a,
  author = {Guo, Hang and Heidemann, John},
  title = {Detecting {ICMP} Rate Limiting in the {Internet}},
  institution = {USC/Information Sciences Institute},
  year = {2017},
  sortdate = {2017-05-01},
  project = {ant, retrofuturebridge, lacrend},
  jsubject = {topology_modeling},
  location = {johnh: pafile},
  number = {ISI-TR-717},
  month = may,
  keywords = {icmp, rate limiting},
  url = {http://www.isi.edu/%7ejohnh/PAPERS/Guo17a.html},
  pdfurl = {http://www.isi.edu/%7ejohnh/PAPERS/Guo17a.pdf},
  abstact = {
  Active probing with ICMP is the center of
  many network measurements, with tools like ping, traceroute, 
  and their derivatives used to map topologies and as
  a precursor for security scanning. However, rate limiting
  of ICMP traffic has long been a concern, since undetected
  rate limiting to ICMP could distort measurements, silently
  creating false conclusions. To settle this concern, we look
  systematically for ICMP rate limiting in the Internet. We
  develop a model for how rate limiting affects probing,
  validate it through controlled testbed experiments, and
  create FADER, a new algorithm that can identify rate
  limiting from user-side traces with minimal requirements
  for new measurement traffic. We validate the accuracy
  of FADER with many different network configurations in
  testbed experiments and show that it almost always detects
  rate limiting. Accuracy is perfect when measurement
  probing ranges from 0 to 60x the rate limit, and almost
  perfect (95\%) with up to 20\% packet loss. The worst
  case for detection is when probing is very fast and blocks
  are very sparse, but even there accuracy remains good
  (measurements 60x the rate limit of a 10\% responsive
  block is correct 65\% of the time). With this confidence,
  we apply our algorithm to a random sample of whole
  Internet, showing that rate limiting exists but that for slow
  probing rates, rate-limiting is very, very rare. For our random
  sample of 40,493 /24 blocks (about 2\% of the responsive
  space), we confirm 6 blocks (0.02\%!) see rate limiting at
  0.39 packets/s per block. We look at higher rates in public
  datasets and suggest that fall-off in responses as rates
  approach 1 packet/s per /24 block (14M packets/s from
  the prober to the whole Internet), is consistent with rate
  limiting. We also show that even very slow probing (0.0001
  packet/s) can encounter rate limiting of NACKs that are
  concentrated at a single router near the prober.
  }
}