LISTSERV mailing list manager LISTSERV 15.5

Help for NORDNOG Archives

NORDNOG Archives

NORDNOG Archives


Next Message | Previous Message
Next in Topic | Previous in Topic
Next by Same Author | Previous by Same Author
Chronologically | Most Recent First
Proportional Font | Monospaced Font


Join or Leave NORDNOG
Reply | Post New Message
Search Archives

Subject: Re: IP-tel
From: Nils Steen <[log in to unmask]>
Reply-To:Network management discussion for Nordic region <[log in to unmask]>
Date:Fri, 31 Jan 2003 00:43:16 +0100

text/plain (67 lines)

On torsdag, jan 30, 2003, at 10:20 Europe/Stockholm, Kurt Erik
Lindqvist wrote:

>>> As we reach higher access-speeds I feel that there is definately room
>>> for
>>> a usage based high-speed service, for instance for ethernet users.
>> Yes, yes!  Tell me when and where I can buy it. I want it both for my
>> company and at home.
>> And, like PAF, I also offer to help calculate a realistic per-bit cost
>> if someone should feel an urge to share usage pattern data. I *really*
>> want good usage-based billing Internet connectivity to happen.
> For this to work though, you need a reliable data gathering method.
> Netflow on high-speeds only do samples, if I remember with 5% deviation
> of actual traffic. Question arises if that is accurate enough to use as
> billing data....

The accuracy increases with the number of samples.

I have a presentation ( ~300K ) which shows ways of post processing
sampled data so as to not over-charge the customer. It also shows that
as the traffic reaches 10E9 bytes you are getting very close to actual
usage with sampled stats.

If there is interest I'll post the pdf or ping me off-list.

> In the IETF the ip-fix WG was supposed to create a protocol that would
> solve this, they are now in the MPLS swap trying to figure out how to
> generate statistics on MPLS sign of intelligent life.
> If I remember correctly, Lasse Hanson (that I think is on the list)
> told me that the problem with high-speed interfaces is not to generate
> the data, it's to get it shipped off the router. So what we would need
> is a protocol that would let you
> a) from a few parameters choose what to generate stats for (source,
> dest, packets, bytes, ports?)
> b) Aggregate it on the router
> c) Send in batch

It is just a question of granularity.
If you limit your usage-based billing to just divide your customers
into high-usage and normal categories, you can get away with using low
sample rates. Usage is commonly billed on a monthly basis so even if
you take few samples, over 30 days this will add up to enough data to
bill the customer without the need to go to great lengths to make your
network cope with the overhead.
On the other hand if you wish to bill the customer per byte,per
destination-as then you must do router aggregation, not to mention some
serious customer education on what all that stuff in their bill
actually is.


> Who writes the draft? :)
> - kurtis -

Back to: Top of Message | Previous Page | Main NORDNOG Page



CataList Email List Search Powered by the LISTSERV Email List Manager