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

Re: Nsfnet and Ebone stats

As to Nsfnet: Seems to me that if one is gonna plunk down millions of
dollars into a federally subsidized network infrastructure to replace
some existing infrastructure - it should at least be upward compatible
in all aspects - which includes statistical monitoring and recording.
If it is impossible to do now - then someone in NSF/ANS/Merit should
come out and say it.

As to Ebone: I realize with ciscos it is impossible to do.  But we
can at least create reports of ifInOctets/ifOutOctets per Ebone backbone
link, as well as ifInErrors/ifOutErrors and perhaps the downTrap to see
how many times a link goes to the down state and for how long.  Data
Communications just had an interesting issue with an article about line
reliability and it showed that on average in Europe line reliability was
around 99.1% as opposed to the CCITT recommendation of 99.6% for
intercontinental links and 99.9% for continental links.  I have been
monitoring my links and find that I average 30 down states a month
sometimes for as short as 8 seconds - and Data Communications goes into
this as a synchronization problem between PTTs - which use different
synch methods - and they can't tell when a link goes down until you call

Here is an extract from an article I have written for a local computer
What are the major causes of a digital telecommunications circuit to
go out of service? Synchronization and multiplexing signals on high
bandwidth international links is more complicated than on slower
speed analog links.  Synchronizing at the international level is
particularly tricky because different carriers (PTTs) often use
different signalling schemes.  These synchronization problems
usually result in a link disruption of under 2 minutes, long enough
to force users off a remote application.

Due to the fact that different PTTs use different signalling standards,
it is close to impossible for PTTs to monitor digital circuits and
instead wait for customers to call them to report circuit outages.
PTTs usually show higher availability statistics than customers, since
they only start their "down clock" when a customer reports the problem,
which is often 15-30 minutes after the line has gone down.

Interestingly enough, the higher the data speed, the more reliable
the digital data circuit (from the INTUG report):

Speed        Uptime
------       ------
768kb        99.99%
384kb        99.94
1544kb (T1)  99.94
1024kb       99.9
128kb        99.88
1984kb       99.88
512kb        99.87
2048kb (E1)  99.56
64kb         99.03
256kb        97

Looking at the above table, users would be recommended to stay away
from 256kb circuits and to upgrade as fast as possible from 64kb to
128kb circuits.  Almost all higher speed lines (other than 64kb and
256kb) meet the CCITT recommendations.

All I am saying is that we need some 'minimal' amount of SNMP recording
and reporting and I am sure we are all doing this on our own national
level so why not expand it to the Ebone?


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