Merit/NSFNET.Information.Services-Interested.Parties(a)um.cc.umich.edu,
Elise Gerich <epg(a)merit.edu>,
Jessica Yu <jyy(a)merit.edu>,
nsfnet-info(a)merit.edu,
nsfnet-reports(a)merit.edu,
smh(a)merit.edu,
"Stefan Fassbender (GMD/EASInet)" <stf(a)easi.net>
In-Reply-To: Message of Wed, 1 Apr 92 13:36:53 EST from <jaw(a)merit.edu>
On Wed, 1 Apr 92 13:36:53 EST you said:
>Olivier,
>
>The use of NNStat for statistics collection on the T1 backbone has
>not been disabled at any point of entry. Ultimately, drastically
>differing in and out counts are not unusual.....................
it may be, but in the case of CERN they are, as CERN is also a data
source.
>................................................................
>Using the net primarily for ftp will appear as a small amount of
>traffic into the backbone and a significantly larger traffic
>volume out.
>................................................................
>We hope this information helps explain your concerns. Please let
>us know if we might provide further clarification.
>
The reason I am very suspicious, is that we are collecting the cisco
accounting frequently enough that we do not lose too much of it, and
we are therefore in a good position to cross-check our results with those
of MERIT.
Whereas, our figures agree with the NSFNet ones regarding the traffic out of
the T1 backbone, there is a complete mismatch regarding the traffic into
the T1 backbone.
For example, MERIT reports 358,200 bytes coming from 128.141, the main CERN
network, whereas we measure 5,239 Gigabytes (factor 14,000). However, we
agree for the traffic to CERN from the US, 5,212 Gigabytes (MERIT),
5,239 Gigabytes (CERN).
Back to EASINet as a whole, I do not think it is realistic to believe that
the real traffic patterns are as distorted as I reported yesterday, namely:
>From EASINet--->NSFNet 78.73 Megabytes (which would be less than the CERN
originated traffic, according to us)
**** YES, this is Megabytes, NOT in Gigabytes ************
>From NSFNet---->EASINet 82.14 Gigabytes (which is probably all right, but
we can only crosscheck part of it
because some of the routers at CERN
are not controlled by us, what we
could measure is: 43.73 GBytes).
To put these figures in perspective, the International Connection Manager (ICM)
which is handling the lines from Cornell to Stokholm (Nordunet) and
INRIA (France) would have, according to MERIT: 68.29 Gigabytes (out) and
43.42 Gigabytes (in)
which is more like the 1/3 2/3 rule of thumb which we observed in 1991, and
which shows that even though the USA are exporting more data than they
receive, the balance is not that bad, especially considering the small
number of FTP servers in Europe, Finland excepted.
Olivier
