◆
Peaks you can predict
Enrolment and term start are known dates. Capacity meets them without being paid for in July.
Institutions have a demand curve unlike anyone else: almost nothing, then everything, on a date known a year in advance. Paying for the peak all year is the traditional answer and a poor one.
Used for student systems, learning platforms, research computing and administration.
What this looks like in practice
Enrolment and term start are known dates. Capacity meets them without being paid for in July.
An analysis gets capacity when the researcher is ready, not when a shared cluster frees up.
An environment that produced a result can be recreated exactly, which is what publication increasingly requires.
A figure that survives contact with the academic year.
Institutional infrastructure is sized for enrolment week and idle for most of the year. The waste is enormous, well understood, and has simply had no alternative.
Enrolment week is the one occasion each year when an institution’s software is judged by everybody simultaneously — students, parents, faculty and press. It is also the week it has historically been least able to cope, for reasons that are entirely about how capacity is bought rather than how the software is written.
Capacity arriving with the question rather than after it has been queued.
The environment behind a published result, recoverable exactly.
The small systems that run real academic work, finally affordable to maintain.
Admissions surfaces that hold up on the deadline day they are built for.
Enrolment week used to define our whole infrastructure budget. Now it is a week.
That is the case this is designed for. Capacity meets a known peak on the date it occurs rather than being carried all year.
It arrives when the analysis does and is released when it does not, so the size of a question stops being set by what is free on a shared cluster.
Yes. The environment is recoverable exactly, which is increasingly what journals and funders expect rather than merely prefer.
It is usually less than the year-round capacity it replaces, because you stop paying continuously for a peak that lasts a week. You will have the figure before the year starts.
Run regulated workloads with a complete record of every change, and answer an auditor in an afternoon rather than a quarter.
Carry sensitive workloads with separation that is structural, and keep clinical and research software moving.
Serve the public on infrastructure that stands up to scrutiny, procurement and a service that cannot be allowed to fail.
Hold up on the day that matters most, and change your storefront in the week you decided to, not the quarter after.
Publish the moment it matters, carry an audience that arrives all at once, and keep an archive that stays reachable.
Bring plant data somewhere useful, keep the software near the line dependable, and stop treating the factory as offline.
Begin with one workload. Keep what you have running. Decide whether to continue once you have seen it carry something real.
No migration required to begin. Nothing to tear down if you stop.