The earlier transactions established how permission and current data move. A fabric must also provide resources for those obligations to finish. Work from a concrete dependency cycle to filtered traffic, then latency, service policies and hierarchical responsibility.
Lesson | After it you can... |
|---|---|
Explain the reachable P/Q resource cycle and the request < forward < response rule that breaks it. | |
Tell protocol deadlock from routing deadlock and check that classes have independent resources. | |
Classify failure using enabled work and useful completion. | |
State assumptions A1-A6 behind a progress claim. | |
Count probes and responses without pretending they are latency. | |
State the inclusive-filter invariant and where false positives come from. | |
Match responses to active obligations rather than raw packet counts. | |
Free an entry only after clean holders ack or dirty data reaches memory. | |
Compute an illustrative critical path for a permission miss. | |
Define eligibility and the scope of round-robin selection. | |
Preserve protocol safety and completion capacity under priority. | |
Find unbounded bypass and reject unsupported cycle bounds. | |
Distinguish global cluster coverage from local cache records. | |
Explain how a cluster's local permissions are bounded by the global record, and count die-crossing legs. | |
Separate link acceptance, coherent completion and failure recovery. | |
Compare total scoped metadata and required paths under stated assumptions. |
Suggested order: resources, filtering, service, then hierarchy. Keep the shared course model open. New finite-resource policies and cluster topology are explicitly introduced where used, rather than silently changing H's ordering or data rules.
It changes the work sent, not necessarily the held/requested resource graph or selection policy. The remaining required data path can still deadlock or be bypassed.