Speaker
Description
Most common CPU architectures provide simultaneous multithreading ( SMT). Thereby, the operating system sees, per physical core, two logical cores and can schedule two processes to one physical CPU core. This overbooking of physical cores enabled a better usage of parallel pipelines and doubled components within a CPU core. On systems with several applications running in parallel, such as batch jobs on worker nodes, the usage of SMT can increase the overall CPU performance.
In HEP, batch/Grid jobs are accounted for in units of single-core jobs. One single-core job is designed to utilize one logical core fully.
Usually, Grid sites configure their worker nodes to provide as many single-core slots as physical or logical CPU cores.
However, several processes running on the same CPU core can also block each other and reduce CPU performance since two processes want to use the same part of a CPU core. Therefore, it can be useful to configure more single-core slots than physical CPU cores but fewer than logical CPU cores per worker node. We have extensively used and studied this strategy at the GridKa Tier 1 center. In this contribution, we show benchmark results for different overbooking factors of physical cores on various CPU models for different HEP workflows and Benchmarks.
| Consider for long presentation | No |
|---|
