Velocity Bike-Share: Is Four Dedicated Hosts Actually Cloud?
Flemish bike-share operator Velocity runs its reservation and payment platform on six physical servers in a leased rack. Average CPU utilisation across the year is 14%, but during the April–September tourist season request volume rises to four times the annual mean, and each of the last two summers the booking front end timed out for several hours on the first sunny weekend. The board has approved a move to a cloud provider, but the CFO has returned the business case with a condition: she will sign only if the solution is 'genuinely elastic and genuinely pay-per-use'. The infrastructure team proposes instead a five-year contract for four dedicated hosts in the provider's data centre, sized for the summer peak, with the same operating system images and the same manual patch process. The architect argues this is 'private cloud' and therefore satisfies the CFO. Meanwhile the marketing team has already bought a SaaS email tool without telling IT.
1. Evaluate the four-dedicated-hosts proposal against the essential characteristics of cloud computing. Is the architect's 'private cloud' argument sound?
Read the worked answer
Test the proposal against each essential characteristic rather than against the label 'private cloud'.
On-demand self-service - fails. Capacity is obtained by negotiating a five-year contract, not by the consumer provisioning unilaterally through a service interface.
Broad network access - passes. The hosts are reachable over the network from anywhere, but this characteristic alone is the weakest discriminator: a 2005 colocation rack also passes it.
Resource pooling - fails. Four hosts are assigned solely to Velocity. Nothing is pooled across customers, and the consumer has no abstraction over where workloads run.
Rapid elasticity - fails. The estate is sized for the peak and cannot grow or shrink without a contract variation, so the summer timeout problem is solved by paying for the peak all year.
Measured service - fails. The charge is a fixed monthly fee per host regardless of consumption, so it fails the CFO's pay-per-use test directly.
Conclusion. The architect is confusing a deployment model with a set of characteristics. 'Private cloud' describes who the infrastructure serves and how it is owned or isolated; it does not exempt a service from the essential characteristics. A genuine private cloud still needs self-service provisioning, pooling and metering, typically via an internal chargeback portal and orchestration layer. As described, this is dedicated hosting with a five-year commitment: it converts a variable demand problem into a fixed cost and leaves the elasticity problem unsolved.
2. The marketing team bought a SaaS email tool on a company card without involving IT. Explain which cloud characteristics that purchase exhibits, and what governance failure it reveals.
Read the worked answer
The purchase is a textbook demonstration of the characteristics the infrastructure proposal lacks.
On-demand self-service is present and is precisely why it happened: a marketing manager obtained capacity in minutes with no procurement cycle.
Measured service is present: it is a per-seat subscription that rises and falls with headcount.
Rapid elasticity is present at the seat level - adding or removing users requires no renegotiation.
Resource pooling is present: the provider runs a multi-tenant service and Velocity neither knows nor cares which physical host serves it.
Governance failure. The very characteristic that makes cloud useful - low-friction self-service - is the one that bypasses procurement, security review and data-protection assessment. The organisation now has a live service it has not assessed for: where mail content is stored and whether that location is acceptable, whether a data processing agreement exists, who can administer the tenant, whether corporate identity is federated or a shared password is in use, what happens to data at termination, and what egress or exit costs apply.
Corrective action. Do not simply ban it; that pushes the behaviour further underground. Publish a short approved-service catalogue with pre-cleared terms, keep a fast path (for example, a two-day review for tools in the catalogue), and reconcile the SaaS spend discovered on corporate cards against the catalogue each quarter. The lesson for the CFO is that elasticity and pay-per-use are properties of the service contract and the provisioning interface, not of where the servers physically sit.