Segment 14
This sub-session focused entirely on analyzing, designing, and specifying a new memory management architecture for the cuzk GPU proving engine. Through a deep forensic audit of the memory lifecycle for 32 GiB PoRep proofs, the assistant and user identified that the existing `working_memory_budget` config was dead code and the static `partition_workers` semaphore was the only throttle, leading to OOM risks. They collaboratively designed a solution featuring a single unified memory budget auto-detected from system RAM, LRU eviction for SRS and PCE caches with a 5-minute minimum idle time, two-phase working memory release, and removal of configurable preload and `partition_workers` in favor of budget-based admission. The final deliverable was the detailed `cuzk-memory-manager.md` specification, defining new core types, evictable caches, and integration points, providing a directly actionable blueprint for implementation.
Designing a Memory-Aware Admission Control System for the cuzk GPU Proving Engine