Estimation of Costs and Pay per Use on a Large-scale Shared Computer Science Testbed: the Grid'5000 Case
Résumé
Grid'5000 is a large-scale science testbed: multiple academic organizations
contribute to some of the costs associated to operating and maintaining it,
to the point many approximations must be made when attempting to build a view
of the overall costs associated to it. However, such an approximated cost has
been estimated, first in 2015, and more recently in 2017 and 2018 as part as
a standardized infrastructure cost evaluation for all large scientific
instruments on France's national research infrastructure roadmap. From this
approximated cost, we can derive pricing.
However, this does not imply that we can make users pay for their usage of
Grid'5000, as:
(1) many users are affiliated in a way or another to the organizations
that are contributing to supporting the costs of the infrastructure;
(2) the way costs were approximated does not qualify Grid'5000 usage as an
eligible cost in the context of funded projects as it is not auditable.
In practice, this only leaves as potential paying users: (1) foreign
academics with no ongoing collaboration with their French colleagues, and (2)
private sector users. Being a minority of the user base, we cannot expect a
significant revenue stream from such categories of users. In theory, we
could also hope to invoice organizations who benefit more than they
contribute to costs, but in practice, such a scheme would be difficult to
operate.
In conclusion, we do not believe that we can transition to a pay-for-use model for a
majority of Grid'5000 users without a major increase in bureaucracy to
associate usage to affiliation and a reduction to the number of organizations
supporting the costs of the scientific instrument. This would certainly have
a negative impact on usage, increasing the number of custom-built
mini-testbeds and therefore the overall cost to the research
ecosystem.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...