<<< Chronological >>> Author Index    Subject Index <<< Threads >>>

Minutes of the Test Traffic working group meeting at Ripe 40.

  • From: "Henk Uijterwaal \(RIPE-NCC\)" < >
  • Date: Wed, 10 Oct 2001 21:34:54 +0200 (CEST)
  • Cc: < >
    < >

---------- Forwarded message ----------
Date: 10 Oct 2001 16:43:28 -0000
From: Keith Godber <keith.godber@localhost

Thanks again to Olaf for taking notes.

Keith

=======================================================


Minutes Test Traffic meeting.
RIPE 40, Prague.
Chair: keith.godber@localhost
Wed Oct  3 09:00  CEST 2001
Minutes olaf@localhost
----------------------------------------------------------------------

A. Administrative

B. TTM, status and plans

Presentation by Henk.

-  Update of RIPE NCC personnel status.
  + Florian Grubers work on site survey using PDA and GPS.

-  Network update
  + Service contracts.
  More action after RIPE 30

   + 14 new TT machines since Bologna
   5 for evaluation
   9 in the chain

   + Less growth than expected

   + Status
   63 in the field: 42 machines taking data.

   + new boxes
   - European Data Grid
   - 3 ISPs
   - Research

   - Glossy
   - Direct Marketing
   - Howto for daily operations
   - Product development.

- Hardware
  + Automatic Connectivity check implemented.
  + Better rsync scheme
  + Motorola clocks replaced for Trimble.
    (Trade possible for EU 800)
  + 1u testbox
    prototype available.
  + GigEth
    Next kernel upgrade will have support
  + 48v power
    Not in 1u will fit in 2 u.
    investigation continues
  + multiple NICs in testbox.
    no fundamental problem to do this.
    control software needs modification.

- Data presentation
  + WWW Interface on the boxes.
  Direct access to local data: OS, satellites and packets.
  available on tt48 (series C) and higher. Older series
  do not have the CPU

- IETF efforts.
  + 4 implementations of OWD and OWL, RFC 2339 2679 2680
  + No interop of measurement devices.
    Unify packet formats and setup mechanisms.
    2 drafts in ippm WG.
    RIPE NCC will implement drafts.

- Analysis

  Summaries and trends
  Bandwidth
  Jitter
  Performance Scores
  2 groups TU-Delft

 + Performance scores
   Physical Distance vs transmission speed does not really work
   TU Delft: Network distance... looks promising

  + Continue with present efforts
    Implements Shavitt paper. (ieee infocom 2001)

  + generate statistical from traceroute
  + Wisconsin univ
- 2002
  + expand measurement network
  + product and service development
  + kernel update
  + User training
  + IETF work
  + Solve N**2 problem. (over 100 boxes.. to much data)

Question: Are most new boxes in EU?
A: No 2 US operators and a global network.


Q: Connectivity measurements.
A: indirect by looking at packet loss. But not implementing RFCs.

C. Summaries and Trends by Wilhelm.

- Motivation
  visual representation has it's limits.

- Solutions
  Summarize data using IPPM owd parameters

  numerical summaries and trend plots.

  Numbers are presented in plots: median and 2.5 and 97.5 percentile
  Color coding for daily variations.

  Trend plots of above data for longer periods. (e.g 3 months)

  Deployment: summaries are in production.
  Trend plots need work.


D. TTM Bandwith studies. Santcroos and Henk Uijterwaal

- Bandwidth
  Motivation
  + curiosity and planning
  IPPM WG charter
  + still under discussion
  + define a BCP
  Requirements
  + little impact on the normal traffic
  + exclude HW and SW effects
  + Real time measurements
- Parameters
  C: bandwidth or capacity
  A: Available bandwidth
  C-A: Utilized bandwidth.

  Start using C.
- Available tools
  RTT tools (pathchar, pchar & clink)
  Packet Dispersion  (pathrate)
  Protocol dependent tools. (Netperf)

- Tools in detail
  (.. the details...)

- Comparison of tools
  pathrate is best in determining the real cable.
  pathrate takes a few times longer to converge.

- Plans
  Continue in pathrate
  Put in testboxes.
  Discus BCP work in IETF context.


Q: Have tests been done with asymmetric links.
A: Pathrate is asymmetric.
   For RTT tools you will not be hurt.

Q: How about complicated topology
A: No results quite yet.

Q: Can you use this for the long distances
A: We can do both, for long pathrate is giving best results.


E. RIPE TT data analysis at AS level by Jan-Pascal van Best.

- Focusing on traceroutes relating to network layout

- Background
  + PhD: Coordination of Internet IP connections
  How inter ISP agreements affct IP QOS
  + UMEEPI project

- Why ASs
  + RIPE TT spans 3600 IP addresses 111 ASs
  ASs reflect routing policy
  + WHOIS data currently used (Route Views would be better?)
  + Only 20 IP addresses unmappable

- Analysis tools
  + Traceroutes ->  IP Graph -> AS graph
  + IMport tool
    Queries TT DB, Whois servers and generate XML
  + Analysis tool
    visualizes the network
    shows evolution in time

  + Available under GPL

- Outlook

 + theoretical analysis of graphs
 + use delay measurements
 + improve graph display mechanism

- Demonstration.

   XML file loaded to generate graph.
     Inter Testbox ASs can be  distinguished from Testbox ASs.
     Size of AS is indicated.
     Possibility to select individual ASs and look at connectivity
     Possibility to zoom into AS and look at time-series within that
     AS.
   Impressive graphical tool.

F. Jitter Analysis by Reinhard Sojka and Henk Uijterwaal

- Motivation
  + Absolute variation does not really matter variations do.
    (VoIP)
  + IPDV metric to measure variations in delay.
    useful if measurement errors in delay are small compared to variation.
  + IP Delay Variations also referred to as jitter.

- IPPM draft.
   by Demchelis and Chimento#

- Draft implemented
  + takes into account experimental errors.
  + plots available
  + writeup available soon

- Demo
  + interesting cases

- Plans
  + get feedback on write up
  + add IPDV to daily plots
  + study interesting cases.

H. Direction of group.


Chair asked for directions of the group

Q: Is netflow work related to the group.
A: Yes it is. Note taken.


I. Call for agenda items for RIPE 41.
 preferably early.

J. AOB
  none.






  • Post To The List:
<<< Chronological >>> Author    Subject <<< Threads >>>