Document Managed by Network Operations

Historically, the cost for networking infrastructure has been offloaded to departments and other customers for the procurement of equipment. In addition, the identified costs were inflated to allow for backbone growth in response to increased demand. This was necessary as demand generally exceeded supply and there were insufficient central resources supplied to allow for growth. Ultimately, this became a flat 7K charge for the allocation of a 64 host network on any router. In addition, some early customer was generally required to pay for all costs associated with the procurement of the router itself. The non-recoverable nature of these costs resulted in general resistance to make these purchases and so larger routers where generally specified to future proof capacity. While these policies provided service to university customers both directly and through the development of the RUNet backbone, it generally penalized early adopters and departments with available resources.

The cost to add a single network to RUNet can be viewed as follows:
• N = O + R + U + I
• N network
• O outside plant
• R router chassis, supervisor, and software
• U uplink
• I network interface

This is very simplistic explanation, but when customers approached OIT with the request to put a building onto RUNet, all of the indicated costs were generated and billed directly. This front loaded all major costs to early adopters and limited the economic exposure of OIT. The addition of other networks to this installed infrastructure at a cost of 7K per network (the "I" component) was a marginalized cost of a single interface and corresponding backbone upgrades to accommodate increased traffic.

The following, different mechanism is proposed to cost out network additions to the RUNet. The outside plant component can be requested of the university directly for any specific building. Where these costs are recovered centrally, they will not be an issue for any occupant. Where they can not be recovered centrally, an occupant will have the choice to absorb these costs directly. Subsequent to deployment, outside plant will be considered a sunk cost to either the university or to an occupant and OIT will not revisit the issue. This represents no change with regard to today.

The router chassis, supervisor, and IOS license will be absorbed by the original occupant. These costs will be recovered identically as part of the original deployment. However, over time, additional customers may utilize these resources and will be required to purchase their portion of this fixed cost by transferring funds to early adopters. For example, customers numbered C1, C2, and C3 each join the network at separate 2 month intervals utilizing the same router. The transfer of funds to provide for the router chassis, supervisor, and IOS would be as follows:

• C1 generates IPO for R+U to OIT TD at month 1
• C2 generates IPO for (R+U)/2 to C1 at month 3
• C3 generates IPO for (R+U)/6 to C1 at month 5
• C3 generates IPO for (R+U)/6 to C2 at month 5

After month 5, each customer (C1, C2, C3) will have a chassis investment equal to (R+U)/3. In this manner, early adopters will recover their chassis investment and all customers will share an equal infrastructure burden. The generic representation is as follows:

This cost is paid out as (n-1) equal IPO's to the (n-1) prior customers.

Early customers are free to waive this infrastructure charge for later customers and allow them to obtain a network for just the interface fee. This would be a departmental matter and OIT would not interfere. However, OIT will take responsibility for notifying the new customer as well as the invested customers of new additions. Deployed infrastructure will continue to remain with the building and vacating tenants will not recover these hardware investments.

The interface cost for a customer will be computed from appropriate averages of the actual hardware costs. This represents the true marginal cost for hardware. As an example, 75XX class routers typically provide interfaces as viper blades and port adapters. A viper can hold 0, 1, or 2 port adapters. However, a viper would not be deployed if there were no requirement to hold a port adapter. In this context, the 0 case will be discarded. Thus, the viper can hold 1 or 2 port adapters and this will occur with equal probability. The average usage for a viper is 3/2 port adapters. Based on this, if the viper has a unit cost of V, then the viper cost per port adapter follows:

viper cost = 2V/3

This is a specific case where the cost is recaptured from equally likely port adapter counts (1 and 2). In general, if there are 1..N equally likely counts, the average count is represented as follows:

average = (N+1)/2

To recover costs, the actual cost will be normalized by the average number of recipients. This provides the 2/3 factor that appeared in the viper cost calculation. This analysis is valid for the port adapter itself. An Ethernet port adapter provides 8 network connections. Thus, it can support 0..8 networks. Removing the lower bound as nonexistent, the PA can support 8 possible networks with each configuration (1..8) occurring with equal probability. Thus, the average usage would be 9/2 networks for an Ethernet port adapter. Thus, if an Ethernet port adapter has cost E, the interface cost for a single Ethernet network using viper and PA8E technology is as follows:

I = 2*(2V/3 + E)/9

A fast Ethernet port adapter provides one network connection. Its average usage would be 1 network. Thus, if a fast Ethernet port adapter has cost F, the interface cost for a single fast Ethernet network using viper and PAFE technology is as follows:

I = 2V/3 + F

In general, OIT will break even on LAN hardware costs in this fashion. There will be losses on some installations and gains on others, but total interface costs should be neutral for OIT. To utilize actual numbers:

• V = 7700 (VIP2-40)
• E = 5600 (PA8E)

Thus, if chassis costs have already been absorbed by an occupant (most common situation), the cost for one additional Ethernet connection is as follows:

I = 2*(2*7700/3 + 5600)/9 = 2,385

An occupant is free to purchase the entire port adapter (which is cheaper in the long run if they expect to occupy more than half) and all interfaces will be marked as reserved for them. If they purchase the entire port adapter, there would be no recovery factor and their cost would include the port adapter itself and the viper charge.

I = 2*7700/3 + 5600 = 10,733

However, this would provide for 8 subnets. Vacating tenants will not recover their interface investment, but it will relocate with them.

In addition, costs associated with a deployment will be provided in detail. Customers are encouraged to review this information carefully and obtain additional information to ensure a comfortable transaction. For more complex projects, we reserve the right to add a contingency fee to cover the procurement of additional, unexpected items (transceivers, patches, etc..). Subsequent to deployment, a detailed listing of these contingency costs will be forwarded to the customer in addition to any refund as appropriate.

In direct contrast to prior conceptions, IP space is free. There will be no charge for IP space, regardless of network size. We do, however, have the responsibility to manage this space effectively, and networks will not be allocated in a fashion that poorly utilize space. Networks of appropriate sizes will be allocated to occupants as requested and under the constraint that the space is properly utilized. Thus, an occupant may have the option of choosing a larger network to avoid additional interface costs. For example, if the occupants have 100 hosts that they would like to be on a single network, a 128 host network may be ideal for them. However, if these hosts are equally split and would benefit from some degree of broadcast containment, than two 64 host networks would be a much better choice. It is our hope that a reduced interface fee (less than 2500) would not pose a significant barrier to making the better networking choice.

Wherever possible, network space will be allocated with an address alignment appropriate to a network that is double the allocated space. A block of addresses will be split into a low block and a high block. The low block will be allocated as the requested network and the high block will be marked as reserved with the intent of allowing an allocated network to double in size without requiring renumbering. All allocations will be performed from free space first, but reserved space will be utilized if there is no appropriate free space to allocate. The inability to grow a network in place will not be a barrier to obtaining appropriately sized address space. However, a larger net may require renumbering if a prior allocation utilized the reserved high block.