Customer oversubscription levels

This might be a dumb question, but I can be sure that I'll be told if that's the case, so here goes:

What's a good oversubscription ratio for customer traffic to global Internet bandwidth these days? I.e., if you have, say 90megs of bandwidth to other transit providers, how much bandwidth, in aggregate, are you selling to customers -- 90? 450? 900?

Do customers care about this? Or do they assume that if they get a T1 to the Internet from you that they have their own T1's worth of non over-subscribed bandwidth to your transit providers?



Got to think most customers assume oversubscription. Having been on the
provider end of this in a previous life, how it often goes is like
this. The customer will think/feel they are not getting what they are
paying for. As a result the customer will deliberately try to peg their
ckt at the bw you say you are selling them, and if they cannot achieve it,
they call you and complain. You at that point need to fix it or risk
losing the customer. As you sell higher and higher bandwidth rates, the
customer's tolerance, presumably because of what they are paying you, goes


I'd look at this slightly differently.. customers wont use all their
available bandwidth at once anyway so you will get aggregation simply by
economy of scale.

Compare to say the PSTN, there might be say 100000 local customer lines
attached to a particular exchange and yet the capacity out onto the trunk
network might be say 5000 lines. Clearly there is a high degree of over
subscription and phone customers are the first to complain when they cant
dial out and yet it doesnt happen because not everyone uses it at once.

The same is true for bandwidth, you can achieve a good level of
aggregation - say 5:1 and yet each customer can max out their own link
because they dont all do it at once

The problem comes when you either just dont have trunk capacity or you try
to put far too many customers onto a node.

I always call it aggregation vs contention - in one you achieve economy of
scale and in the other you knowingly degrade service (within the bounds of
an SLA)

So, to answer your question, it depends what your selling as to how much
oversubscription you can achieve, anything for 3:1 to 20:1 depends on if
its transit, leased line, broadband, dialup... but I think you should look
more at whether you are actually degrading an individuals service or just
achieving aggregation.


Typically Class 4 have an over subscription ratio of 4:1, the biggest Class
4 switches can handle 40,000 DS0s like the AT&T 5Ess or DMS 500 switches.

A over subscription ratio of 100000/5000= 20 is too high for PSTN.

FYI modem to user ratios are 10:1 on the average.

"Stephen J. Wilcox" wrote:

Yeah sorry should have included a disclaimer! My numbers for PSTN were
made up as I dont have any real figures to hand or in my head .. it was
more the analogy I wanted to demonstrate as people seem to understand
aggregation in local exchanges better than bandwidth contention even tho
its the same really..

Ok real life.. How about we have about 200 extensions on our PBX but only
30 lines onto the provider network and yet we never see engaged...


By now, I think it's widely accepted that it really isn't
"oversubscription" or "overselling" until congestion starts becoming an
issue. Up until then it's "statistical multiplexing".