An ECN Probe-Based Connection Acceptance Control. Kelly, T. Computer Communication Review, 31(3):14--25, July, 2001.
@article{ Kelly01,
  author = {T. Kelly},
  title = {An {ECN} Probe-Based Connection Acceptance Control},
  journal = {Computer Communication Review},
  year = {2001},
  volume = {31},
  number = {3},
  pages = {14--25},
  month = {July},
  annote = {In this paper, an ECN (Explicit Congestion Notification) probe-based admission control protocol for realtime non-adaptive traffic is presented. Briefly, the admission control protocol works as follows. Suppose a host A wishes to send a realtime stream to a host B with a peak rate, R, and a packet size, s. Host A will initiate the realtime stream if the proportion of probe packets that experienced congestion is less than a threshold level, eps. The probing phase starts with host A sending a packet to host B requesting B to open a so-called probe stream. Throughout the probe phase, host A calculates the current proportion of acknowledged packets that experienced congestion, p_total. When host A updates p_total, if p_total > eps + 1/n then host A stops to send probe packets, and the realtime stream is not admitted. Upon acknowledgement of the first probe packet, the probing takes place in rounds of length approximately one round-trip time. During each round, host A calculates the proportion of packets which experienced congestion, p_last. Initially, the probe packet is sent at a rate of s/RTT. This rate is then increased by 2 - p_last for each consecutive round. When the target rate R has been reached, the final round is entered. The probing terminates when host A has entered the final round and received acknowledgements for at least 1/eps probing packets. The performance of the admission control protocol in terms of delay and loss is simulated using ns-2. Three scenarios are considered: only non-adaptive traffic, multi-hop bottlenecks, and an integrated network, i.e. a network with both non-adaptive and adaptive (TCP) flows. The outcome of the simulation experiment suggests that the admission control protocol works well in the non-adaptive scenario and in the integrated scenario as long as the load is moderate. However, since the TCP flows in the integrated scenario do not use the admission control mechanism, they efficiently block out the non-adaptive flows at high traffic loads. As is to be expected, the blocking probability for multi-hop connections in the multi-hop scenario have higher blocking probability than single-hop connections. However, compared to per-hop measurement-based admission control schemes, the admission control protocol is less discriminate against multi-hop connections, i.e. it has a higher admission probability than the product of the per-hop admission probabilities.},
  bibdate = {Wednesday, May 22, 2002 at 10:36:32 (CEST)},
  submitter = {Karl-Johan Grinnemo}

Downloads: 0