End-to-End Available Bandwidth: Measurement Methodology, Dynamics, and Relation with TCP Throughput

1
End-to-End Available Bandwidth Measurement
Methodology, Dynamics, and Relation with TCP
Throughput

• M Jain, C Dovrolis
• Proceedings of ACM SIGCOMM02, 2002
• Presented by Ivan

2
Outline

• Introduction
• Self-Loading Periodic Streams
• Measurement Tool Pathload
• Verification And Discussion
• Comment

3
Available Bandwidth

• Available bandwidth of the end-to-end path during
  a time interval

Determine the end-to-end available bandwidth
Determine the end-to-end available capacity
Figure1. A pipe model with fluid traffic for a
three-hop network path
4
Available Bandwidth (cont.)

• The end-to-end avail-bw is defined as the maximum
  rate that the path P can provide to a flow,
  without reducing the rate of the rest of the
  traffic in P.
• Narrow Link ? the link with the minimum capacity
• Tight Link ? the link with minimum avail-bw

5
Outline

• Introduction
• Self-Loading Periodic Streams
• Measurement Tool Pathload
• Verification And Discussion
• Comment

6
Self-Loading Periodic Streams (SLoPS)

• A periodic stream in SLoPS consists of K packets
  of size L, sent to the path at a constant rate R.
• If the stream rate R is higher than the avail-bw
  A, the one-way delays of successive packets at
  the receiver show an increasing trend.
• Use the increasing delays property in an
  iterative algorithm to measure end-to-end
  avail-bw.

7
Self-Loading Periodic Streams (SLoPS) (cont.)

• Suppose SND sends a periodic stream of K packets
  to RCV at a rate R0. Denotes the One-Way Delay
  (OWD) DK.
• The OWD difference between two successive packets
  k and k1 is
• ,
• Proposition 1.
• If R0 gt A, then gt 0 for k 1,K-1. Else,
  if R0 ? A, 0 for k1,K-1

8
An iterative algorithm to measure A

• R(n1) can be computed as follows,
• Rmin and Rmax are lower and upper bounds for the
  avail-bw after stream n.
• Initially, Rmin0, and Rmax is a sufficiently
  high value gt A.
• The algorithm terminates when
• Rmax Rmin ? ? (estimation resolution)

9
SLoPS with real cross traffic
Grey-Region
10
Outline

• Introduction
• Self-Loading Periodic Streams
• Measurement Tool Pathload
• Verification And Discussion
• Comment

11
Some issues

• Clock and timing issues
• Not affect ?Measure relative OWD DK
• Stream parameters
• A stream consists of K packets of size L, with
  rate R.
• Packet inter-spacing T, packet size LRT, stream
  duration VKT

12
Detecting an increasing OWD trend

• Suppose the OWDs are D1, D2, , DK, partition
  into G groups, then compute the median OWD
  of each group
• Pairwise Comparison Test (PCT)
• ,
• PCT measures the fraction of consecutive OWD
  pairs that are increasing. 0? SPCT ?1.
• OWDs indep ? SPCT ?0.5
• Strong increasing trend ? SPCT approaches 1

13
Detecting an increasing OWD trend (cont.)

• Pairwise Difference Test (PDT)
• PDT quantifies how strong is the start-to-end OWD
  variation. -1? SPDT?1.
• OWDs indep. ? SPDT ?0
• Strong increasing trend ? SPDT approaches 1
• Default, SPCT gt0.55, SPDT gt0.4 ? Increasing

14
Fleets of Streams
Limit the average pathload rate to less than 10
of R
Total Duration N(V?)

• To determine whether RgtA, it sends a fleet of N
  streams.
• The average pathload rate during a fleet of R ?

15
Grey-Region

• Using a fraction f of N streams to determine
  type-I (increasing) or type-N (non-increasing)
• Grey-region ( )
• Less than Nf streams are increasing trend
• Less than Nf streams are non-increasing trend
• Grey-region upper bound ? Gmax
• Grey-region lower bound ? Gmin
• Default f is set to 70

16
Rate Adjustment Algorithm
Increasing trend ( R(n) gt A ) Rmax
R(n) R(n1) (Gmax Rmax)/2 Non-increasing
trend ( R(n) lt A ) Rmin R(n) R(n1) (Gmax
Rmin)/2 Grey region R(n) gt Gmax Gmax
R(n) R(n1) (Gmax Rmax )/2 Grey region
R(n) lt Gmin Gmin R(n) R(n1) (Gmin Rmin
)/2
Grey region
Terminate if Rmax Rmin lt ? (estimation
resolution)
or Gmax Gmin lt ? (grey-region resolution)
More detail in Pathload a measurement tool for
end-to-end available bandwidth
17
Measurement Latency

• Default K100, L800B, T100µsec, a stream
  carries 80,000 bytes and lasts for 10msec
• If A ? 100Mbps and ? 100ms, pathload takes less
  than 15 sec to produce a final estimate
• The latency increases as the avail-bw or
  grey-region increase, and it also depends on ?
  ?

18
Outline

• Introduction
• Self-Loading Periodic Streams
• Measurement Tool Pathload
• Verification And Discussion
• Comment

19
Accuracy Results
MRTG range is 6Mbps
MRTG range is 1.5Mbps
Fall within the MRTG range in 10 out of the 12
runs
Fall within the MRTG range in 9 out of the 12 runs
20
Variability and Load conditions

• Relative variation
• The variability of the avail-bw increases
  significantly as the utilization u of the tight
  link increases

21
The Effect of the Stream Length Fleet Length
Fig13. The effect of N on the variability
of the avail-bw.
Fig12. The effect of K on the variability
of the avail-bw.
Fleet duration increases ?Variability of the
avail-bw increases, Variation across
different pathload runs decreases
Stream duration increases ?Variability of the
avail-bw decreases
22
Is Pathload Intrusive?
Avail-bw measurements
RTT measurements
Do not show a measurable decrease when pathload
runs.
23
TCP vs. Available Bandwidth
RTT measurements
Avail-bw and TCP Throughput

• TCP connection manages to saturate the path
• Shorter TCP connections expect a significant
  variability in their throughput
• TCP connection will grab part of the throughput
  of other TCP connections

24
Outline

• Introduction
• Self-Loading Periodic Streams
• Measurement Tool Pathload
• Verification And Discussion
• Comment

25
Comment

• Pathload is the first generation tool using the
  SLoPS method to measure the avail-bw
• IGI pathChirp are the modified ones
• The similar measurement accuracy
• With less bw-usage overhead
• With less measurement latency

26
Pathload vs. pathChirp
From the pathChirp Efficient Available
Bandwidth Estimation for Network Paths