Cost Recovery and Rates
The HPSCC serves a heterogeneous population of users ranging from students with modest needs to power users with special purpose computing and storage needs. We provide an open, fair and transparent financial model that encourages sharing of resources while minimizing the effective "taxation" of users to fund the research efforts of power users.
The HPSCC operates on the basis of what we like to call a "post-modern condo model." This is a variation of a powerful and popular "condo" resource sharing strategy. The "condo" strategy is in use at other academic institutions. The basic idea is quite simple and equitable. Briefly, Labs, Departments, Institutes and schools, (basically whomever needs computing resources), buys their own resources and then, instead of managing the resources themselves, they place their resources in a common location and share the costs of operating the resources (i.e. electricity, HVAC, system administration, licenses, etc.). All the operating costs are shared between the groups in proportion to the amount of hardware/software that each group has under management. This way the labs and departments reduce costs, and increase stability, and can afford greater system administration expertise. The result is a cost-effective computing environment operated by very experienced sysAdmins and one faculty. Many groups have recognized the value proposition and become stakeholders by placing their hardware under HPSCC management. This has allowed significant growth and produced a large computing environment that has become a resource for SPH and SOM. The HPSCC "owns" about 10% of the resources for sharing by everyone. No one stakeholder currently owns more than 30% of the resources.
The "post-modern" part of the model is how we handle users who do not contribute hardware. This is handled differently at different institutions. At many institutions they are allowed to use excess capacity for free, but at a lower priority. The HPSCC handles non-stakeholders by charging them at exactly the same rate as the stakeholders. This defrays the costs of the stakeholders, thereby providing them incentive to share their hardware. This is how we are able to provide cost-effective service to everyone: Stakeholders and non-stakeholder's alike. It also spreads the costs somewhat over a larger base of users in order to provide greater stability for the organization.
Our cost recovery algorithms calculate the
net monthly cost for each resource by adding the base cost of each
resource (e.g. the monthly depreciated purchase cost, upgrades, etc.)
to a fraction of the common costs (e.g. salaries, HVAC
service contracts, etc.). All users who use a particular resource in a
given month, share the net monthly cost of that resource in proportion
to their monthly usage of that resource. This cost is the same whether the user is a stakeholder of the resource or not.
Unlike most service
centers where the resources are few and fixed, the HPSCC currently has
over 40 resources and over 100 users across four departments. Many of
these resources are upgraded at unpredictable times in the course of
each year in response to faculty requests. In such an environment,
effective planning for the purposes of determining rates, presents an
intractable problem. To sidestep this problem we take a novel approach
to determining rates: The monthly net cost and hence the effective
rate for each resource is not fixed over one year, but can fluctuate
from month to month in response to financial events, e.g. upgrades, end
of depreciation periods, or the addition of new resources to the
Allowing rates to fluctuate from month-to-month allows
the HPSCC to dynamically adjust resources in response to the business
and research opportunities of the faculty, while minimizing adverse
billing impacts on regular users (e.g. users being effectively taxed
to support the research of power users). The entire monthly cost is completely recovered on a monthly basis. The fluctuations do not present a problem for planning, since fluctuations are typically only a few percent.
Charges are calculated monthly, but billed
quarterly for collection. Since each of our
hundred or so users typically submits hundreds of jobs to each of our 40 or
so resources, it is not possible to manually determine the monthly
charges. Instead, all calculations are performed automatically by an
accounting application written by the HPSCC staff. This application
extracts usage and cost data stored in a MySQL database and calculates
cumulative monthly charges for each user. All users are charged the same rate per unit of fractional
usage of the same resource. There are no special rates or subsidies for special
users. The accounting system is blind to a userâ€™s
institutional affiliation. Bills are sent to department administrators and include usage reports broken out by
user id, resource id and department or budget.
A spreadsheet showing the monthly charge of 100% usage of our resources from 2009 can be downloaded here. Since most resources are currently fully depreciated or have been paid-for by specific users, the costs largely reflect the salaries of system administration staff, service contracts and supplies. As of Feb. 2011 our computing charges are approximately $38/core-month while our storage costs average (~$0.20/GB-month). The actual costs vary from device to device.