SCIC meeting de-brief
Les briefly covered the main relevant points from the SCIC meeting on 7/6/99.
There were several presentations on recent upgrades or expected for the CERN,
IN2P3, Japanese (see
Update on
Japan-EU), Italian
(see
http://www.ba.infn.it/~ruggieri/Presentazioni/ICFA-SCIC6Jul199/index.htm),
Canadian (see
Canada Report) and German international links.
Dean Karlen
of Canada
reported that the performance between Canada & DFN is abysmal. A letter is being
drafted by Michael Ernst to send to the DFN. A second letter has been drafted
related to the KEK - European link to send the appropriate agencies in Japan and
Europe.
The satellite provider between DESY & the FSU has changed and they hope for
better performance.
There has been very little activity from the working groups so far.
They want to do a technology update and
provide trends, Harvey Newman, Richard Mount & Micahel Ernst are on this WG.
They want to provide
some expectations on how networks will evolve, there may be some interesting data
from the recent CERN tender.
Les Cottrell made a presentation on the network monitoring (see
Status of SCIC Working Group on Network Monitoring).
The organizers of the Lepton Photon Conference '99, have asked Matthias Kassemman
to make a 15 minute presentation on the
SCIC activities. He will want 2-3 slides from the network monitoring people. he will
put together a summary of what he is going to talk about, a week before the
conference (9 Aug-99) and will circulate for comments.
The next SCIC meeting will be at CERN on Saturday November 3rd, 1999.
We brainstormed on what should go into the 3 transparencies for Matthias. The following ideas
were put forward:
- Architecture diagram with sites.
- Deployment map.
- Map with with Beacon links colored based on performance.
- Long term graphs of performance.
- Collaborations and their performance.
- How is the pingER data used.
- Comparisons of different network providers & peering.
- Problem areas.
- Futures.
This is more than will fit on 3 transparencies, so Les will go away an come up with a
distillation.
PingER projects
SAS efforts at HEPNRC
We agreed to stick with the SAS database/analysis package at FNAL for another year.
Michael will be going to SAS training this week. HEPNRC will look into purchasing SAS/ETS
(for time series analysis) and SAS/STAT. They will also look at the SAS ODBC package for
Unix which allow remote SQL access to the SAS database. Warren is working with CNRI to use
the posgress database which is SQL com,pliant so hopefully the work could use SAS later.
Bill is also looking at splitting off the archive onto its own machine which could help
response time.
When we get up to speed on SAS then it might be rewarding to think of using SAS for:
-
Production of
other graphs, e.g. frequency histograms, Fourier analysis (possibly), other
metrics (TCP thruput prediction, variability) other ways of showing data
(Interquartile range, box plots, percentiles etc.)
- Using SAS to aggregate the data, e.g. into daily, weekly, monthly bins and
also providing the summarized results in Excel format.
Generating alerts
We talked in the past about generating alerts from the data, e.g. by making
rolling baseline measurements and comparing the current measurements to
their probability and after some number are out of range raising alerts. Bill is going
to look at whether he can hire a student to tackle this project. He will know something in 3
weeks.
Current SLAC projects
Other things SLAC is working on include:
- Thruput vs loss & RTT correlation (using Mathis et. al. formula).
- Comparing AMP data to PingER.
- Understanding how the various active meaurement performance projects such as
PingER, Surveyor, AMP and RIPE compare (for a start see
Comparison of some Internet Active End-to-end Performance Measurement Projects).
- Building an improved top level report (i.e. like a statistically
defensible internettrafficreport. We will probably start from the
PingER Summary
Tables.
- Poisson scheduling, recording more information for each ping (i.e. the
result of each ping so can go back later and use different statistical
measures, e.g. 90%ile etc.), allow flexibity in the packet sizes and
frequency
(in particular to be more friendly to poor links such as S. America).
Bill thought this might be a
good project for Michael, and Warren is willing to let Michael take it over.
- Looking at Ping jitter vs VoIP jitter and also the effect of QoS.
- We also have a student working on Java DB connectivity and using Swing for
generating graphs.
- Creating PingER monitoring pages for various collaborations including
BaBar, CDF, D0, RHIC etc. For an example see
Network
Monitoring for the BaBar Collaboration.
Ongoing work at Oxford
John MacAllister was working on porting traceping to perl. We have
requested an update, but
have heard nothing for a couple of months. XIWT has added
traceroute monitoing to their version of PingER and if we don't hear
something
from John soon, maybe we should go ahead and do something without John.
Les will continue to try and make contact with John.
Wishlist items
Some other things on the wishlist include:
- Looking for ICMP rate limiting signatures (e.g. loss(100bytes) <<
loss(1000byte), loss on later pings in a series >:> loss on early pings in
a series, loss on pings that need
fragmentation (e.g. over 1460 bytes MSS), comparing ping with http.
- Look some more at measuring the performance of applications such as HTTP
(e.g. deploy the Intel
timeit
code, measure DNS etc). Bill said he would look at this.
- Comparing the performance of the commodity internet & R&E nets.
Next Meeting
This will be a phone meeting on Wednesday 28th July at 2pm PST.
Feedback
