� � What do Packet Dispersion Techniques Measure? Constantinos Dovrolis, Univ-Delaware Parmesh Ramanathan, Univ-Wisconsin David Moore, CAIDA � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 1 of 28
� � Overview � Background: capacity and available bandwidth � Dispersion of packet-pairs � Dispersion of packet-trains � A capacity estimation methodology: pathrate � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 2 of 28
� � Part I Background � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 3 of 28
� � Definition of capacity � Maximum IP-layer throughput that a flow can get, without any cross traffic Link-1 Link-2 Link-3 A C Source Sink � � � � C i ( i � � � � H ) i : capacity of link � Path capacity C is limited by narrow link n : � min f C g � C C i n i �� ��� H � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 4 of 28
� � Definition of available bandwidth � Maximum IP-layer throughput that a flow can get, given (stationary) cross traffic Link-1 Link-2 Link-3 A C Source Sink � u i i : utilization of link � Available bandwidth A limited by tight link t : � min �� � � � �� � � A C u C u i i t t i �� ��� H � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 5 of 28
� � Part II Packet-pair dispersion � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 6 of 28
� � Packet-pair: basic idea L � Packet transmission time: � � C � Send two packets back-to-back � Measure dispersion � at receiver L � Estimate C as � ∆ L/3C =L/C L/C 3C C � But.. cross traffic ‘noise’ can affect the packet dispersion � � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 7 of 28
� � Previous works on packet-pair dispersion � Largely considered cross traffic effects as ‘random noise’ � Carter and Crovella (Infocom 1997), Lai and Baker (Infocom 1999): Statistical techniques to extract most common measurement range (mode) � Paxson (Sigcomm 1997): observed multimodalities but did not relate them with cross traffic � They suggest use of maximum-size packet-pair packets � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 8 of 28
� � Multimodality of packet-pair estimates � Cross-traffic causes local modes below (SCDR) and above (PNCM) capacity mode (CM) P={100,75,55,40,60,80}, L=Lc=1500B P={100,75,55,40,60,80}, L=Lc=1500B 400 160 SCDR 360 Capacity Mode (CM) 140 # of measurements # of measurements 320 120 280 u=80% 100 240 u=20% Sub−Capacity CM Dispersion Range 200 80 (SCDR) 160 Post−Narrow 60 Capacity Mode 120 40 (PNCM) 80 PNCM 20 40 0 0 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 Bandwidth (Mbps) Bandwidth (Mbps) � � � Heavier cross traffic load makes CM weaker Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 9 of 28
� � Creation of SCDR and PNCM modes � SCDR is caused by cross traffic interfering with packet-pair L/60 L/60 L/60 1 CT 2 C =60Mbps t i-1 L/40 L/40 1 2 C =40Mbps t i L/40 + (2 L/60 - L/40) = L/30 � PNCMs are caused by back-to-back packet-pairs after narrow link (first packet is adequately delayed) � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 10 of 28
� � Effect of cross traffic packet size � Distinct cross traffic packet sizes cause SCDR local modes � Common Internet traffic packet sizes: 40B, 550B, 1500B P={100,75,55,40,60,80}, u=50%, L=1500B P={100,75,55,40,60,80}, u=50%, L=770B 160 120 Fixed CT packet size: Lc=1500B Variable CT packet size: 110 140 Lc uniform in [40,1500]B SCDR 100 CM # of measurements # of measurements 120 90 SCDR 80 100 70 80 60 CM 50 60 PNCMs PNCM 40 40 30 20 20 10 0 0 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 Bandwidth (Mbps) Bandwidth (Mbps) � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 11 of 28
� � Effect of packet-pair size � Previous work suggests use of maximum-sized packets � But.. this is not optimal for uncovering capacity mode P={100,75,55,40,60,80}, u=50% P={100,75,55,40,60,80}, u=50% 120 220 Lc : uniform in [40,1500]B 110 200 Lc : uniform in [40,1500]B 100 SCDR 180 # of measurements PNCM # of measurements 90 160 CM 80 140 70 L=100B 120 L=1500B 60 100 50 80 CM 40 SCDR 60 30 40 20 20 10 0 0 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 Bandwidth (Mbps) Bandwidth (Mbps) � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 12 of 28
� � Packet-pair dispersion: summary � Packet-pair technique: simple in unloaded paths � Multimodal bandwidth distribution in loaded paths � Most common measurement range (mode) is not always the capacity � Capacity is normally a local mode (CM) � Interfering cross traffic packets cause local modes or SCDR � Loaded post-narrow links also cause local modes (PNCMs) � Maximum packet size is not optimal for uncovering CM � How can we tell which local mode is related to the capacity? � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 13 of 28
� � Part III Packet-train dispersion � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 14 of 28
� � Packet-train dispersion � What do we measure with the dispersion of packet trains? ∆ (N) ∆ (N) ∆ (N) 2 R 0 3 N 2 1 3 2 1 3 2 1 N N S C C C R 1 2 3 CT CT CT � N � �� L � Bandwidth estimate: �� N � � What is the effect of length N on bandwidth estimate? � Carter & Crovella: packet-train dispersion estimates A � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 15 of 28
� � Packet-train experiments � What happens as we increase the packet-train length N ? P={100,75,55,40,60,80}, u=80%, L=Lc=1500B P={100,75,55,40,60,80}, u=80%, L=Lc=1500B 200 140 N=3 N=2 180 120 160 # of measurements # of measurements 140 100 120 CM 80 100 60 80 60 40 40 CM 20 20 0 0 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 Bandwidth (Mbps) Bandwidth (Mbps) � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 16 of 28
� � Packet-train experiments (cont’) � Range of measurements decreases and becomes unimodal � Measurements tend to Asymptotic Dispersion Rate (ADR) (less than C ) P={100,75,55,40,60,80}, u=80%, L=Lc=1500B P={100,75,55,40,60,80}, u=80%, L=Lc=1500B 400 240 N=5 Asymptotic 360 N=10 Dispersion 320 # of measurements 200 # of measurements Rate R=15Mbps 280 160 240 200 120 160 80 120 80 40 40 0 0 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 Bandwidth (Mbps) Bandwidth (Mbps) � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 17 of 28
� � ADR in single-hop paths � For sufficiently large N : � N � �� L C � R � � � C � � C � � � u � � � C � � ADR is not the capacity C � ADR is not the available bandwidth A � For single-hop paths, we can estimate A from R C � � � � � � A C � R � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 18 of 28
� � Effect of cross traffic routing � Path-persistent and one-hop persistent cross traffic Path Path Source Sink 1 2 3 H . . . C C C C C 0 1 2 3 H Cross Traffic Cross Traffic Sources Sink Cross Traffic Sinks Path Path Source Sink 1 2 3 H . . . C C C C C 0 2 1 3 H Cross Traffic Sources � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 19 of 28
� � Packet-train dispersion in multi-hop paths � Can we derive the ADR for general cross traffic routing and paths? � Cross traffic packets cause ‘bubbles’ between packet-train packets � For one-hop path persistent cross traffic with � C C : i C C � R � H � � max u � � P u i �� ��� H i i i �� � Lower bound: bubbles are never filled in � Upper bound: bubbles are determined by tight link � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 20 of 28
� � Packet-train dispersion: summary � Packet-trains: do not lead to more robust capacity estimation � Packet-trains: do not lead to available bandwidth estimation � As N increases, measurement range decreases and becomes unimodal � As N increases, measurements tend to ADR � ADR is always less than capacity � Available bandwidth can be computed from ADR in single-hop paths � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 21 of 28
� � Part IV A capacity estimation methodology � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 22 of 28
� � Pathrate : a capacity estimation methodology Phase I: � Perform many (2000) packet-pair experiments to form distribution B � Use packet size of about 800 bytes (maximum size: 1500 bytes) � Determine local modes of distribution B � Sequence of local modes in increasing order: M � f m g � m � � � � m � � M � � Constantinos Dovrolis - IEEE Infocom 2001 - April 24-26, 2001 23 of 28
Recommend
More recommend
