packet triplet: an enhanced packet pair probing for path capacity estimation
DESCRIPTION
Packet Triplet: An Enhanced Packet Pair Probing for Path Capacity Estimation. Nanyang Technological University Zou ZiXuan. OUTLINE. The purpose of Packet Triplet Previous Packet-Pair based techniques Packet Triplet probing Detailed approach Filtering scheme Experimental Results. - PowerPoint PPT PresentationTRANSCRIPT
Packet Triplet: An Enhanced Packet Pair Probing for Path
Capacity Estimation
Nanyang Technological University Zou ZiXuan
OUTLINE
The purpose of Packet TripletPrevious Packet-Pair based techniquesPacket Triplet probingDetailed approachFiltering schemeExperimental Results
The purpose of Packet Triplet
Measuring the path capacity (bottleneck)Packet-Pair basedRobust to cross traffic
Previous Packet-Pair based technique
Packet-Pair dispersion Originated by Keshav for congestion
control Cross traffic effect—expression and
extension Variants of Packet-Pair to deal with the
effects of cross traffic.
Previous Packet-Pair based technique
Statistical method Sending many packet pairs, using
statistical approach Union and Intersection Kernel density
Do not always lead to correct estimation Not a global modal but the local modal Observed by Paxson, investigated by Dovrolis
Previous Packet-Pair based technique
Model based detection Based on Multimodal distribution Packet-Train and Packet-Pair
Packet-Triplet probing
Deal with multimodal distributionAccurately estimate path capacity in different load traffic Different from model-based detection
Detailed approach
Three back-to-back probing packets and dispersion pair
3 2 1
L LL
3 2 1
recei ver si de
1 2
Sender si de
12
12
21
jj
jj
jjj LBB
11 / LB 22 / LB
Detailed approachPacket Triplet probing
Idea condition No cross traffic
The variation of Latency of probing packets
Equivalent estimates
3 2 1
L LL narrow l i nk i
3 2 1
i i
recei ver si de
02
21
i
iijjj LBB
012
21
jj
iijjj LBB
),( 21jj BB
j
Detailed approachCross traffic effect When
When Narrow link capacity estimates (NLE)
j
3 2
123
1
1j2
j
1j2
j
j
3 2 1
L LL
3 2 1
recei ver si deSender si de
1 i 2 i
narrow l i nk i
ii CL /
21
),( 21jj BB
Cont’d Post-narrow link estimates (PNLE)
Equivalent noise (EN) Occur in low frequency
),( 21jj BB
kk CL /3 2 1
L LL post-narrow l i nk j
3 2 1
recei ver si de
i i 1 k 2 k
cross traffi c
21
),( 21jj BB
kij
kij
and
and
2
1
3 2 1
1j 2
j
Cross traffi c di stort i on
path
Filtering scheme
1. filter out the estimates in case of
as the noises. 2. Of the remaining estimates, the estimate of lowest value and relatively
high frequency of occurrence is selected as the set of NLE. The mean of the NLE estimates is considered as the bottleneck of measured path.
),( 21jj BB
j
),( 21jj BB
Experimental ResultsSimulation setup ns-2 Topology
P={50,20,10,30,25,40} Mps 1000 packet triplet experiments are done.
Mbps5.0
Cont’dPacket-Pair vs. Packet-Triplet L=100bytes cross traffic load=20%,
50%, 80%
Cont’d L=700bytes CL=20%, 80%
Cont’d L=1000 bytes CL=20%, 80%
Conclusion
The noise filtering and bottleneck detection are always efficient whenever the cross traffic is light or heavy when we choose the small probing packet size how to set is a remaining issue for our research on this topic
Thank You !!!!!!