Over-Subscription

Frame Relay operates on the concept of sharing.  Multiple users share the same resources.  This is the only reason why Frame Relay is so much cheaper than Private Lines, which are dedicated (one Private Line per customer, and nobody can share that circuit).

Similar to a highway - drivers share the roads.  Private Lines allow each driver to have their own private road . . . oh sure, they are guaranteed no congestion, but the inefficiency of it just doesn't make sense.  Oversubscription is the same as designing a highway that can handle 10,000 drivers simultaneously - to allow for 1,000,000 drivers per day.  They bank on the fact that the drivers NEVER all use the road at the same time.  Similarly, Frame Relay customer almost NEVER all use their circuits at the same time, and even if they do, the NEVER EVER all max out their connections simultaneously.

Since the provider charges less for Frame Relay, they must be able to oversubscribe - sign up more users than their network could handle simultaneously.  All Frame Relay providers oversubscribe !!!

They rely on several facts, based on the theory of statistics and porbability:

• not all customers are using their Frame Relay circuits simultaneously
• those who are using the services, are sending bursty data - the data rate goes up and down
• statistically, the more PVC's, the less likely it is that all PVC's will be transmitting a high data rate at the same time

Example

Imagine that you are a provider, and at one edge switch, rated at 5 Mbps - you have signed up 10 customers, each with one PVC that is allowed a maximum of a T1 data rate (1.536 Mbps).  Imagine also that at any instant in time, that all are using their PVC's.  Now, if all 10 customers are, at a given instant, maxing out their PVC and sending 1.536 Mbps, then the total data rate would be 15.36 Mbps !!  The switch would choke, and reject much of the traffic.

BUT - this never, ever happens !!  Data networks, always transmit bursty data.  There are occasions when they may max out the PVC (file transfers, for example), but this is a rarity, and when this does happen, the switch employs a fairness algorithm so that they can only send data at that rate - if the capacity of every switch along that path allows it.

If we look at two customer PVC's - here is a typical data transmission chart for ach:

As you can see, the additive bandwidth does not max out at any given instant for both - and even if it did, it would never max out for all 10 customers.

But it Could, Right ??

Well yes, but - lets say, since the data rate is fairly random, that we have 10 people flipping coins - heads is full data rate of 1.536 Mbps and tails is a low data rate of 0.3 Mbps.  As they flips the pennies over and over again, once in a great while all will flip heads.  When this happens, frames are dropped.  But the customer then resends the frame and it will get through on the next attempt.  Also, since this occurence is very rare it is not a problem.

How much Oversubscription??

A general rule of thumb for edge switches is 200%.  For core switches it is much higher.

Taking our previous example, but with core switches.  The core switches have thousands of PVC's going through them.  So image 1000 people flipping a coin.  Now, what is the probability that all 1000 will flip heads at the same time ??  Basically, zero !!!

The key is - the more PVC's, the greater the allowable oversubscription.  So for core switches, the rule of thumb is 500% oversubscription.  So, of a switch maxes out at 10 Mbps, then you could configure it with PVC's whose max ratings total 50 Mbps.

This concept is the key to the success of Frame Relay.  Without oversubscription, there is no advantage to "sharing" of resources.  Oversubscription allows providers to get more out of their network.