Cost Targets for Trusted Servers

[itaysharfi]

Hi WICG members,

We are seeking feedback on cost objectives. We are considering the following plan:

1. Establish reference cost targets based on official Cloud pricing metrics.
2. Publish benchmark frameworks to aid in AdTechs number generation.

We are considering establishing a system where the target is that AdTechs pay $X for every million requests made.

While the exact benchmark details are not yet ready for discussion, the cost targets are open for discussion. Therefore, we would like your input on how to set targets. We are considering specifying a cost per million remarketing requests, but we are open to alternative suggestions as long as they meet the following criteria:

1. Measurable using benchmarks
2. Generic enough to apply across AdTechs
3. Take into account existing supported technologies (e.g., supported clouds, availability of accelerators, etc.
4. Consider realistic Protected Audience models, budding logic, KV and use cases
5. Upper bound estimates so we can use them to define production acceptance as we continue to improve cost

If you have standard benchmarks or stats we should rely on in our evaluation would also be great.

We are also interested in feedback regarding non functional requirements of the tested system such as:

1. Acceptable latency
2. Redundancy
3. What is important to be counted towards cost (egress, ingress, compute)
4. Load variance - average/peak QPS, etc.

We plan to discuss this topic in the next WICG but welcome your thoughts in advance.

Best regards,
Itay