In principle, it might be possible for there to be "quality of service" tiers to a given client.
The simplest scenario that the author can reason through is to order requests for execution in the order of those that have the least execution budget remaining.
In this way, workloads can self-indicate if they're fine for a given request to take a longer period of time.
In practicality, it'd have to be such that the workload supplied either a timestamp or (more likely) a delta which was converted by the application into a timestamp (offset against fixed point), at which point it would be a case of consuming the queue in-order but inserting it to the appropriate spot intelligently.
In principle, it might be possible for there to be "quality of service" tiers to a given client.
The simplest scenario that the author can reason through is to order requests for execution in the order of those that have the least execution budget remaining.
In this way, workloads can self-indicate if they're fine for a given request to take a longer period of time.
In practicality, it'd have to be such that the workload supplied either a timestamp or (more likely) a delta which was converted by the application into a timestamp (offset against fixed point), at which point it would be a case of consuming the queue in-order but inserting it to the appropriate spot intelligently.