Oversubscription: Balancing risk and benefit

Last time, we began a discussion about how to determine the appropriate level of oversubscription when configuring your frame relay services. On the one hand, any time you oversubscribe a service, you run some risk of network congestion. However, if you don't oversubscribe the service, you lose a critical economic benefit and should consider using leased lines.

In our example, we described a hypothetical topology with roughly a 4-1 ratio of oversubscription. The configuration contained 100 56K bit/sec client-side ports; each connected to permanent virtual circuits (PVC) with a 48K bit/sec committed information rate (CIR). The PVCs connected the 100 sites in a hub-and-spoke topology to a T-1 (1.5M bit/sec) port at the host site. Question: Is that central-site T-1 dangerously oversubscribed in this scenario?

We don't think so, for most applications. In fact, we consider it a conservative configuration. Here's why.

1) If you assume that the application has symmetrical traffic, the odds are low that traffic is flowing in a substantial amount in both directions at the same time, even though the service will support full-duplex connections with simultaneous traffic in both directions. Consequently, this means that the oversubscription is effectively about a 2-1 ratio. But, if there were indeed an asymmetrical flow, then these calculations would have to be based on an analysis of the actual flows.

2) In many cases, the CIR is based on the speed needed for appropriate application response time rather than the average traffic flow. In other words, there are two criteria for choosing the CIR. One is the total number of bits transported across the network; the other is how fast you want those bits transmitted. A high CIR permits really fast response times, even though the percentage of time that the line is in use and actually transmitting bits may be rather small. Thus, in many cases, the CIR may be based more on the desired response than the actual number of bits transported, thus resulting in a lower percentage of time that the CIR is actually used - and making a 4-1 ratio a safe choice.

3) In the example we discussed, a large number of low-speed circuits were served by the high-speed port connection. And with any statistically multiplexed network, the larger the number of low-speed connections, the higher the allowable concentration. For instance, if the connections involved at the remote sites were all at T-1 speed, then only one connection could saturate the host T-1 port. However, since the connections were at roughly 5% of the port speed, it would take simultaneous transmissions from 25 of the sites to saturate the port.

In the next newsletter: Oversubscription from the carrier perspective

---

Steven Taylor, consultant and broadband packet evangelist, and Joanie Wexler, an independent networking technology editor and writer, team up to bring you this analysis and commentary. Taylor specializes in education and market analysis, and Wexler adds incisive reporting and research. For more detailed information on most of the topics discussed in this newsletter, connect to www.webtorials.com, the first Web site dedicated exclusively to market studies and technology tutorials in the Broadband Packet areas of Frame Relay, ATM, and IP. Feedback and additional topic ideas are welcome. Please contact taylor@webtorials.com or joanie_wexler@mindspring.com.

Cost analysis of oversubscription Network World, 06/28/99.

Why oversubscribe, anyway?
Network World, 06/23/99.

More benefits of oversubscription Network World, 06/21/99.

"Playing the averages" with oversubscription Network World, 06/07/99.

Archive of Network World on Frame Relay newsletters