Internet Measurement Methods

1
Internet Measurement Methods

• Workshop on QoS
• Hanoch Levy
• Feb 2004

2
The objective

• A talks to B.
• A wants to know how well it goes.
• How well it will go?

3
Performance Measures

• Delay
• The time it takes a packet to go from A to B.
• Loss
• Will the packet arrive or not?
• What fraction of packets will get lost.
• Jitter
• What is the variability of the delay?
• Bandwidth
• At what rate can I transfer my bits to the
  destination?

4
What are typical performance measures

• Look at http//special.matrixnetsystems.com/ratin
  gs/index.asp
• Round trip delay 40ms 700 ms
• Loss 0 - 12
• Goes down on terrestrial links, yet high on
  wireless
• Jitter 10ms, 30 ms 100 ms.
• Bandwidth depends.

5
The causes for problems
6
The causes for problems

• Sources for delay Packets have to
• Traverse links Propagation delay
• Function of distance roughly speed of light,
  0.1-0.2 sec around the globe
• Be transmitted Transmission delay
• Line rate / packet size
• Wait on line Queueing delay
• Number of packets line rate / packet size
• Sources for loss
• Queue is full
• Noisy line (quite uncommon today except for
  wireless)

7
Who is Who references

• IETF the standardization body of Internet.
• IPPM IP Performance Metrics a working group of
  measurements under IETF.
• http//www.ietf.org/html.charters/ippm-charter.htm
  l
• In there find drafts for measuring
• Delay
• Loss
• Delay variation (jitter).

8
References (2)

• Conferences
• INFOCOM
• SIGCOMM
• SIGMETRICS

9
Performance Measures and applications (1)

• Applications differ in the quality measures they
  require
• Voice
• Send a packet every 30 ms.
• Packets must arrive at real time (lt200 msec)
• If dont arrive makes no sense!
• If arrive irregularly confuse recipient!
• Lost packets are OK if not often.
• ? major quality factors
• Delay
• Loss
• Jitter

10
Applications (2) FTP

• FTP File transfer protocol
• Want the file to transfer as soon as possible
• Packet delay not important
• Packet Loss Directly not important
• Packet jitter not important
• Bandwidth Important! (loss indirectly).

11
Applications (3)

• Web
• Want your page to make it in a few seconds
• SEMI real time.
• Network delay is not major factor (since it is
  anyhow less than a second).
• Losses -- a factor ? can slow down the transfer
• Bandwidth (how much bandwidth is available)

12
Bandwidth Measurement

• The route to destination consists of routers and
  links.
• The delay incurred on the path
• Where q_i queueing delay
• l_i latency
• S/b_i transmission delay

13
Bandwidth Measurement

• Path Capacity bandwidth (PCB)
• Min_i (b_i)
• Interpretation If the pipes were in my hands
  how much could I push?
• Path Available Bandwidth (PAB)
• Min (b_i-c_i) where c_i is the cross traffic.
  
• Interpretation how much BW can (will) I get.

14
Bandwidth Measurement (3) Packet pair

• Send a packet pair (two packets back to back) on
  the route
• Property the arrival time difference is equal to
  the propagation delay over the bottleneck (S.
  Keshav 91 )

15
Bandwidth Measurement (4) Packet pair
Explanation
 
16
Bandwidth Measurement (5) Packet pair
Conditions

• The routers on the path are store and forward
• The two packets are sent sufficiently close to
  one another
• Both packets take the same route to the
  destination.
• There are no multi-channel links.
• There is no cross traffic!

17
Bandwidth Measurement (5) Packet pair

• If no cross traffic Measures PCB
• If there is cross traffic and routers use Fair
  Queueing ? measures PAB
• Fair Queueing
• Will explain in the sequel
• More or less divides the channel to pieces
  where each application gets it share of the
  channel.

18
Bandwidth Measurement What we want on the
project

• Want to get the route where one can push the data
  as much as possible.
• Want the route with the highest Available BW
• Better a pipe of 100MByte with 30Mbyte avail than
  200MB with 20 avail

 
19
Bandwidth Measurement

• Carter and Mark E. Crovella 96
• Bprobe pairs of packets sent for a roundtrip
• Use echo packets (ICMP) by the sender
• Send several pairs
• Use histograms to clean those affected by cross
  traffic
• Measures PCB
• Measures round-trip BW ?
• Does not need receiver cooperation ?

 
20
A histogram CC
21
Another histogram
22
Bandwidth Measurement

• Carter and Mark E. Crovella 96
• Cprobe Send a train of packets
• Measures PAB (avail BW)
• Again packets are ICMP
• ? Does not require target cooperation ?
• ? Measures only the round trip BW ?
• Train length used several trains of 8 packets
• Do ICMP represent real traffic? ?
• (low priority at routers)

 
23
Bandwidth Measurement Train accuracy

•  
• Examine the train value against the BW of a
  real stream

24
Bandwidth Measurement Issues (1)

• How large should the train be?
• Too small train may not represent reality
• Too large
• Waist ammunition
• May affect the BW on the route (if very large!)
• How frequently should we send it?
• Too low frequency What you see now is not what
  will happen later (traffic on network changes)
• Too high frequency
• Waist ammunition
• May affect the BW on the route (if very large!)

25
Bandwidth Measurement Issues (2)

• Should we use ICMP or regular packets?
• ICMP A special IP packet that when sent to
  destination, destination returns it right away.
• Used to implement ping.
• Advantage Can be used for measurement without
  cooperation of other side.
• Disadvantages
• Round-trip and not one-way.
• Does not reflect the REAL TRAFFIC
• Routers on the route may treat ICMP differently
• Give low priority
• Drop

26
Bandwidth Measurement Issues (3)

• Should we measure the real transaction?
• Reflects the real traffic best
• If data is not used at other end then we
  wasting data.
• If we measure the real transaction (FTP, say)
• What kind of window should we use to estimate the
  BW.

27
In case you have Questions

• Ippm_at_advanced.org
• hi, i'm writing from italy, my name is david
• i saw you wrote
• 
  
• gt By the way, I have a question ...
• gt What's the difference between a bottleneck b/w
  and an available b/w?
• gt Which of b/w's can "pathchar" give me?
• The available bandwidth along a path between two
  Internet hosts is
• equal to the bandwidth at the bottleneck. In
  other words, in the
• context of pathchar, they're the same thing.
• 
  
• yes, but pathchar doen't work with crosstraffic.
  the vps(variable packet
• size) works
• with a empty link, don't it?

 
28
References (1)

• Internetworking with TCP/IP By Douglas E. Comer,
  Prentice Hall 1995
• Van Jacobson. Congestion avoidance and control.
  In SIGCOMM 88 Conference Proceedings, pages
  314-329, Stanford, CA, USA, August, 1998.
• Van Jacobson. Pathchar - a tool to infer
  characteristics of Internet paths. Presented at
  the Mathematical Sciences Research Institute
  (MSRI) Slides available from ftp//ftp.ee.lbl.gov
  /pathchar/, April 1997.
• S. Keshav - "Congestion Control in Computer
  Networks". Ph.D Dissertation.Department of EECS
  at UC Berkeley, August 1991.
• Vern Paxson. Measurements and Analysis of
  End-to-End Internet Dynamics. Ph.D. dissertation,
  University of California, Berkeley, April 1997.
•  

29
References (2)

• Robert L. Carter and Mark E. Crovella. Measuring
  bottleneck link speed in packet-switched
  networks. Technical Report TR-96-006, Boston
  University Computer Science Department, Boston,
  MA, USA, March 1996.
• Constantinos Dovrolis, Parameswaran Ramanathan,
  David Moore What Do Packet Dispersion Techniques
  Measure In Proceedings of IEEE INFOCOM 2001
  http//www.cs.utk.edu/dunigan/pktprb.ps.
• Kevin Lai, Mary Baker, Nettimer A Tool for
  Measuring Bottleneck Link Bandwidth, In
  Proceedings of the USENIX Symposium on Internet
  Technologies and Systems March 2001
• Kevin Lai and Mary Baker. Measuring link
  bandwidth using a deterministic model of packet
  delay. In SIGCOMM 2000 Conference Proceedings,
  Stockholm, Sweden, August 28-September 01, 2000.

30
References (3)

• Allen B. Downey. Using pathchar to estimate
  Internet link characteristics. In SIGCOMM 99
  Conference Proceedings, pages 241250, Cambridge,
  MA, USA, August 31September 3, 1999. ACM SIGCOMM
  Computer Communication Review, 29(4).
• W. Jiang, T. F. Williams, Detecting and measuring
  Asymmetric Links in IP Network, Tech Rep,
  CUCS-009-99, Columbia University, 1999
  http//www.cs.columbia.edu/wenyu/papers/asym-gi99
  -ea.ps.
• J. Postel. Internet control message protocol.
  Request for Comments (Standard) 792, Internet
  Engineering Task Force, September 1981.
• http//bandwidthplace.com/speedtest/