Wide Area Bandwidth Questions

By Joe St Sauver(joe@oregon.uoregon.edu)

This article attempts to answer some common questions about the capacity and performance of OWEN/NERO, the consortia that provides Internet public connectivity for higher education, public K12 and State of Oregon agencies.

One of the most common questions we hear is, "Is the OWEN network currently operating at its peak capacity?" This is a tricky question, because OWEN adjusts the size of its Internet transit capacity to satisfy observed peak inbound capacity requirements. OWEN also monitors (and will adjust) its internal intrastate backbone circuits if required, although those circuits currently have sufficient reserve capacity to meet forecast needs.

Our Transit Bandwidth Profile

OWEN/NERO's current UUNet transit bandwidth profile as of late January 2000 looked like this:

The tall peaks represent inbound (Internet-to-us) circuit utilization, and the lower line represents outbound (us-to-Internet) circuit utilization. The horizontal axis represents time, with each spike being roughly a day apart. The vertical axis is traffic measured in Mbps (megabits per second). This is a very typical graph for OWEN/NERO transit connectivity, and we can learn some important things from it:

• Use tends to peak dramatically during the day, with the greatest peaks occurring during early to mid- afternoon. Use also exhibits periodicity associated with the day of the week.
• Generally speaking, inbound use dominates outbound use, and inbound use determines the amount of transit bandwidth we require (inbound and outbound capacity can't be independently provisioned at different levels).
• Usage is not "flat topping." If we were grossly underprovisioned, usage would typically exhibit a characteristic "flat top/plateau" area leading up to, through, and immediately after the period of peak demand.

We're confident that we're obtaining the actual capacity we've purchased from UUNet, and that we haven't "overpurchased" capacity we don't currently require to meet peak demand.

Sizing our transit circuits to meet inbound peaking loads is clearly the key. If we buy more capacity than we need, we'll spend more than we should--and for no benefit (we can't "warehouse" excess network capacity; if we can't use it, it's simply lost...a wasted expenditure). If we buy less capacity than we need, particularly less than we need to meet peaking loads, the network will perform poorly and our customers will become dissatisfied.

"How about establishing a bandwidth standard per user, and provisioning bandwidth accordingly?"

People sometimes suggest provisioning our bandwidth according to a per-use standard. The theory behind this is that such a standard would tend to cap (lower) OWEN/NERO's bandwidth requirements. Once you "run the numbers," it becomes clear why we don't have such a standard:

OPEN (one of OWEN/NERO's customers) has 530,000 students. We know that OUS has just over 60,000 students. Let's assume there are roughly 10,000 DAS agency employees who have access to OWEN/NERO (we believe this number is grossly low; however, for the sake of argument and for the convenience of using round numbers, let's go with that estimate). Totaling that up, OWEN/NERO services roughly 600,000 users.

OWEN/NERO has 121 Mbps worth of commodity transit bandwidth, 76 Mbps from UUNet and 45 Mbps from Cable and Wireless (CWIX). If you divide the 121,000,000 bits per second of bandwidth by our estimate of OWEN/NERO's 600,000 users, that works out to:

121,000,000 bits per second
	= roughly 200 bits per second per user
600,000 users

Note that dialup modem connects users at 56,000 bits per second ("56 Kbps"). Clearly, 200 bits per second is a very modest level of commodity bandwidth by any standard, far below what would potentially be required if we were to guarantee even modem level throughput to all OWEN/NERO customers.But what if we were to approach this problem in reverse, computing the committed bandwidth we'd need to provision if we were to guarantee every user minimum simultaneous access at modem-like speeds of 56,000 bits per second? Doing the math for that, we obtain:

(600,000 users) * (56,000 bps) + 33,600,000,000 bps or 33,600 Mbps

That's 277 times the current OWEN/NERO bandwidth--financially unfeasible and technically unrealistic.

Finally, suppose we ask, "What would OWEN/NERO's budget be if every person serviced by OWEN/NERO paid a dollar a month* for their Internet connectivity?" The number is a rather staggering $7,200,000/year, far more than the roughly $1,500,000-$2,000,000/year that OWEN/NERO currently spends.

Bottom line? OWEN/NERO delivers world class connectivity at a bargain price.

*Note: AOL and most other Internet service providers offer service at $15-$20/user/month; however, they offer a variety of value-added services over and above the network access that OWEN/NERO provides. We chose the dollar/user/month figure as a value representing only the network connectivity part of a typical user's monthly Internet costs--although obviously that number could vary for a given user or provider.