Research · 2024–

Letting cloud resources clear like a market

Auctus is a bidding protocol for cloud allocation — pricing contention instead of queueing it.

Fallstudien sind auf Englisch verfasst.

Problem

Allocation under contention is usually answered with queues and static quotas — mechanisms that hide the real question. When demand exceeds supply, something is being valued; the system just refuses to say what.

Ansatz

Let workloads express value and let the allocator clear. Auctus treats contention as a market-clearing problem, with the protocol — not an operator — deciding who runs when supply tightens.

Stand

A paper on this work is under review. Specifics — venue, results, figures — stay offline until the process concludes. That's how review is supposed to work.

The noticing.

Cloud platforms answer scarcity with the politest possible fictions: priority classes, static quotas, retry queues. Each one is an implicit statement about which workload matters more — made once, by an operator, long before the moment of contention.

But contention is a pricing problem wearing an infrastructure costume. When demand exceeds supply, value is being assigned whether or not the system admits it. Auctus starts from the position that the assignment should be explicit, expressed by the workloads themselves, and settled by protocol.

Approach.

Workloads carry bids — a declaration of what a unit of resource is worth to them, now. The allocator clears the market instead of draining a queue. The interesting problems live exactly where you'd expect: fairness under adversarial bidding, starvation resistance, and whether clearing can stay cheap enough to sit in the hot path.

Where it stands.

A paper on this work is currently under review. Double-blind review means the details — venue, title, results, figures — stay off this page until the process concludes. Not because the work is fragile, but because the process deserves the same respect the work does.

If a result only holds when nobody checks it properly, it isn't a result.

lab notebook, on why review matters

Weiter

The review cycle runs its course. The open questions — fair clearing under adversarial load, allocation as a first-class protocol concern — are longer than any single paper.