Draft Minutes of RIPE34 TT-WG

[RIPE NCC Meeting Registration <]

Minutes RIPE34 TestTraffic Working Group (1st draft)

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

Chair: Matthew Robinson

Scribe: Rene Wilhelm

Attendees: 38

 
Agenda: 

A. Administrative stuff 

- appointment of scribe

- agenda bashing

- review of action items

B. Presentation by Henk Uijterwaal on current project and future plans.

C. How the data gathered by the TT project might be further analysed?

D. Publishing policy for collected data.

Z. AOB

 

-----------------------------------------------------------------------

B. Presentation by Henk on current project and future plans

[slides are online at http://www.ripe.net/test-traffic/Talks/9909_ripe34]

Manpower

Johann Gutauer joined on August 6

- Austrian student, at the NCC for 7 months

- working on data-analysis in particular trends in the data

Test-box installation

On 16-Sep-1999 43 test-boxes:

- 31 operational

- 10 in the installation / problem solving phase

- 1 at remote site, installation yet to be started

- 1 spare at RIPE NCC

Maps with test-box locations [see slides]

Proposed change in managing DNS entries for test-boxes:

- NCC will insert A-records for ttXY.ripe.net names
  and also a LOC(ation) record once the box is active

- Hosts can keep existing entries OR insert CNAME records

Future series of test-boxes

- Motorola will stop production of the VP-Oncore [the GPS receiver 
  used in the current test-boxes] soon; replacement unit exists but
  not pin compatible. There is interest in the user community to redisgn 
  the interface board.

- one or two other candidate GPS units; these too require changes
  in the interface.

- Following developments, we don't want to make a decision right now

QUESTION: Have you considered using DCF77 

Henk: Yes, but this is not available in all of the RIPE NCC service

area(*) and requires lots of seasonal corrections

(*) check for example http://mbr-ch.com/EuroUhren/euromap.jpg [RW]

 

Test-traffic in 2000

RIPE NCC Activity Plan for 2000 (to be approved by the Annual General

Meeting of the membership on 19 October 1999)

- In 2000 TestTraffic will become a "membership service" 
  that one can buy from the NCC

- buy as many boxes as you like (and we can produce/support) once 

- become a member and pay a yearly fee for operating them

- Boxes can also be bought for "private" experiments; in that
  case we will provide basic installation support only.

Test-box operation under membership service:

- essentially the same as today: 

- "black" box, measure network between host-site an other test-boxes

- Hosts get access to the data, daily plots and
  all new products based on the TT-data.

The NCC will draft a proposal for the charging by January 2000,

which will then be discussed with the TT-hosts and working group.

By 2001 TestTraffic should be a self sustained project.

Analysis and New services

Data Quality Monitoring 

- Started with regular checks of Antenna conditions and daily plots

- Already found (and solved) a few problems

- Will automate and expand as we go along

Analysis Framework

- Hardware sufficient to keep one year of data online

- Data in ROOT: easier, more flexible analysis environment

- Developed tools to extract subsets of the data for TB hosts

Daily Plots

- hosts asked if they could show daily plots to their customers
for service leves agreement (SLA). Three options:

1. NO, password just for the test-box host

2. YES, always

3. In Between, two passwords: one for testbox hosts, one for customers

The first would be the same for all boxes (allowing hosts access to
all plots), the second would differ per box and could be used to
give customers access to plots from their provider only.

Consensus seems to be option 3

Delay alarms [see slides for illustration/example]

- Inform testbox hosts of significant changes in the delay
(not regular day/night effects)

- Use Long-term-average, Short-term-average (LTA-STA) algorithm.

- prototype to find optimal parameters

- now running on one testbox for several weeks: typically 5 alarm
  messages per day (e-mail), most of them real

- time scale for further deployment of this system: weeks

- needs feedback from tt-hosts once installed

QUESTION: would it be possible to send alarms through the
syslog mechanism instead of via e-mail? 

Henk: yes, no major obstacle; e-mail notification was easiest
to implement, we also plan to add notification via SNMP traps.
The delay alarm software notifies the operators of both sending and
receiving testboxes. It was noted that in case of a major problem local
to one site, alarm messages will go out to most other sites, but
those cannot do anything to fix it. Although this is less desirable,
it seems difficult to avoid: the software runs locally on every testbox,
but judging the potential relevance of an alarm to a host requires 
a larger view of the measurement network.

Loss alarms

- usefull as well, but have the problem that data cannot arrive for 
  two reasons: connection lost OR no data sent by remote box

- needs more thought

Future work

- Long Term Trends

- Qualitiy of the Network overview page

- Effects of packet size on delays

- N2 problem (which paths to measure with increasing number of boxes)

- Continuation of of comparison of results from several groups

- Topics brought up in the IPPM working group of IETF

- Your suggestions ... 


C. How the data gathered by the TT project might be further analysed?

Matthew presented a list of ideas, which triggered some discussion.

- people create own graphs

- share experience, create a library of "graph" code 

- ROOT being the application of choice for datastorage, it may
  be worthwile to setup a repository of macros to create various graphs

- A quick poll of the audience learned that none of the testbox hosts
  present had the ROOT software installed

- Delays vs. Users Online vs. Time

- need to collect more data

- combine with other data: traffic levels, size of routing table, ...

- Delays vs. Destination vs. Time

- single graph showing the 'hot' destination as day goes by.


D. Publishing policy for collected data.

Current Policy: only test-box hosts get access to the data

Options: 

- Continue current policy

- Unrestricted free access

- NCC TestTraffic group publish reports

- restricted use by the public

Three categories of data:

- all data from all boxes

- data from individual boxes

- aggregate data

After some discussion it was agreed to

- create/publish summary data first 
  (daily plots could easily be misinterpreted by outsiders)

- decide on more detailed data later 

What should be included in the summary?

- numbers: average, median, etc.

- summary over all plots

Name hosts or just locations? 

- explicitly naming hosts has the danger of "league tables" being made,
  marketing departments might jump in

- geographic locations on the other have only limited value 
  in the Internet topology

- RIPE NCC will work on drafts, examples of what it would look like

Z. AOB

Due to personal circumstances, Matthew will not be able to chair
the TT-WG session at the next RIPE meeting (21-25 February 2000)
He asks for a one-time replacement.