perfsonar development work

[Nicolas Simar <>]

FYI - input for next week discussion about the metrics...

-------- Original Message --------
Subject: [GN2-JRA1] APM meeting feedback
Date: Fri, 24 Nov 2006 14:51:29 +0000
From: Nicolas Simar 
<>
To: GN-JRA1-list 
<>
CC: '' <>,	Eric Boyd 
<>, Joe Metzger <>

Hi,

I just finished my presentation at the APM meeting.
The demo went well. Thanks to all of you who have take part to it :-)
Presentation in attachment.

I think it went well and had some interests as I didn't find people
sleepy (and it was just after lunch!)

Those are the comments received.
# BWCTL MP
- Possibility of setting-up the MTU when starting a test to find out the
largest MTU possible along the path. Useful based on past experience
(Jerome Durand - Renater).
- Concern about the tests taking the bandwidth of other application.
Otto Kreiter (Dante) suggested that the tests should go through LBE so
not impacting the BE traffic. Harvard Eidnes (Nordunet) answered that
this is not a problem as is is a TCP flow which shares equally the
bandwidth with the other flows. It was stressed that it wouldn't be good
to have UDP tests at any speed. This should be limited.
- Thomas Schimdt (DFN) mentioned that it would be helpful, to interpret
the tests results, to have additional information about the tests, in
particular the limitations as the interfaces capacity to which the box
is connected (fastEthernet, GigabitEthernet or 10GE).
He also raised the point that the TCP throughput is mostly dependent on
the delay between the two servers. So that information would be helpful.
He also mentioned that it would be great if we could have as point of
comparison a theoretical calculation of the expected throughput based on
the delay and of the buffer (which ones?). So they got a point of
comparison.
Niels Den Otter (Surfnet) mentioned that also having as point of
comparison the history of the measurement (or an average, min, max) when
available would be good.
- Harvard Eidnes (Nordunet) wondered if mbps was milli bps or Mega bps.
(m should be replace by M) ;-) .
- Harvard made the point that end-users don't necessarily badly
configure their TCP stack, but rather used the default configuration
which is not suitable for large data transfer.
- We experienced with an error during the demo. The second time it
worked fine. (See remark in the General section on the information about
the availability of the service)

# ABW
- Question from Thomas Schmidt, can we calculate with packet capture
card some information at an average of 10ms (the average buffer time)?

# L2 status (Willi Huber)
- CRC errors are important to measure degradation of the service. (more
than 1 or 2 CRC errors per minutes, then it is time to act bcs there is
a degradation. Sometimes months without any CRC.
- Otto mentioned that for transmission, no degradation are to be
expected. It worked or not.
- Thomas mentioned that degradation can appear.
- The loopback on the transponder are a good way of isolating the fault.
- Otto mentioned that for the switches, he doesn't know if there will be
a problem.

# perfsonarUI
- the audience was tensed and suspended to my lips when presenting the
traceroute feature :-)
- Otto asked about why we had to select those networks. Would be better
without. I mentioned that it is a question of time, until the LSes are
there.
- Otto was asking about the scalability of the system in regards to the
whole of internet (40,000 MAs).

# No specific comments about looking glass or CNM, but a very good interest.

# General
- Question about the information about availability of the services.
They would like to know in advance if a box is available, rather than
starting a tests and wait for 30secs and get an error. Such a
functionality would allow them to trust more on the infrastructure and
avoid frustration when being answered with an error.
- Jerome Durand mentioned that another category could be added (NREN
technical staff, in the case the NREN doesn't run the NOC).
- Harvard Eidnes (Nordunet) asked if it would be possible to also show
input errors and the various sub-counters for the prototype. (relevant
to psUI and CNM).

Action:
- NS to put the hands out on the wiki and a do not redistribute statement.


--
Nicolas
______________________________________________________________________

Nicolas Simar
Network Engineer

DANTE - www.dante.net

Tel - BE: +32 (0) 4 366 93 49
Tel - UK: +44 (0)1223 371 300
Mobile: +44 (0) 7740 176 883

City House, 126-130 Hills Road
Cambridge CB2 1PQ
UK
_____________________________________________________________________


        
                                        


--
Nicolas
______________________________________________________________________

Nicolas Simar
Network Engineer

DANTE - www.dante.net

Tel - BE: +32 (0) 4 366 93 49
Tel - UK: +44 (0)1223 371 300
Mobile: +44 (0) 7740 176 883

City House, 126-130 Hills Road
Cambridge CB2 1PQ
UK
_____________________________________________________________________

Attachment: NS-06-122v3 - perfSONAR for APMs.ppt
Description: MS-Powerpoint presentation