hubcio opened a new issue, #2387: URL: https://github.com/apache/iggy/issues/2387
Use crate https://github.com/HadrienG2/hwlocality to improve CPU affinity so that shards are aware of NUMA boundaries (avoid hyperthreading contention). Current State - Basic CPU affinity in `core/server/src/configs/sharding.rs` via `CpuAllocation` enum - Simple ranges/counts without NUMA topology awareness - Uses `compio::runtime::RuntimeBuilder::thread_affinity()` in `bootstrap.rs:148` Implementation 1. Detect NUMA topology - identify nodes, physical/logical cores, memory localities 2. Extend `CpuAllocation`: ```rust NumaAware { nodes: Vec<usize>, // Which NUMA nodes to use cores_per_node: usize, // Cores per node avoid_hyperthreads: bool, // Skip SMT siblings } ``` 3. Shard distribution - map shards to NUMA nodes, pin threads to domains 4. Memory locality - allocate shard buffers on local NUMA node -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
