Linux 7.3 Improves Cluster Load Balancing for Intel Hybrid CPUs

The patch series developed by Intel engineer Ricardo Neri fundamentally improves load balancing in cluster-aware scheduling for Intel Core (Ultra) series hybrid CPUs that mix P-cores and E-cores (LPE cores). According to a report by Phoronix’s Michael Larabel, these patches are scheduled to be merged during the Linux 7.3 merge window and have already been queued in the sched/core branch of tip/tip.git.

The Challenge of Cluster Scheduling

Cluster-aware scheduling is a feature introduced into the Linux kernel in 2021. It was designed to efficiently distribute task load across clusters of CPU cores that share L2 cache and other mid-level resources. The design assumes a uniform topology and works appropriately on groups of cores with equal performance.

The problem arises with Intel’s hybrid architecture. Processors starting with Alder Lake feature a configuration in which high-performance P-cores and high-efficiency E-cores are mounted on the same chip, creating asymmetry in performance capacity among cores. Cluster-aware scheduling did not function correctly on topologies with such asymmetric capacity.

Neri points out the following in the patch description:

Cluster scheduling aims to maximize performance by spreading load across clusters of CPUs that share mid-level resources. It works well on uniform systems, but it breaks down on topologies with big and small cores arranged in clusters. As a result, it fails on several generations of Intel processors already shipped and upcoming.

In a partially busy system—that is, a state where there are idle CPUs and each busy CPU has one task—the scheduler assigns non-fitting tasks to big cores. However, the remaining tasks are supposed to be spread across the small cores. When CONFIG_SCHED_CLUSTER is enabled, these remaining tasks are expected to be evenly distributed among the clusters of small cores. In the current implementation, this even distribution was not achieved.

Specific Details of the Existing Problems

Neri’s patch series simultaneously resolves multiple issues that hinder cluster-level load balancing on systems with asymmetric CPU capacity. According to the patch descriptions, the details, rationale, and code changes for each issue are explained within individual patches.

Tests were conducted on three generations: Alder Lake, Lunar Lake, and Panther Lake. Alder Lake has clusters of P-cores and E-cores with SMT, and was verified under both SMT enabled and disabled conditions. On Lunar Lake and Panther Lake, tests were performed on a different topology in which E-core clusters are not connected to the L3 cache. The same experiments were repeated with CONFIG_SCHED_CLUSTER disabled, confirming that the load balancer exhibited the expected behavior.

Phoronix is reported to plan benchmark testing using Intel Core Ultra processors once the Linux 7.3 cycle begins, measuring the performance impact across various workloads. No performance figures have been published for the patches themselves.

Historical Background and Scope of Impact

Since the introduction of cluster-aware scheduling, Phoronix has pointed out regression issues on hybrid CPUs. Some time after its introduction, regressions in scheduling were reported, and improvements were made in subsequent kernel releases, but the state remained unoptimized.

As related articles, our site has reported on the effects of cache-aware scheduling on MySQL workloads in Linux Cache Aware Scheduling Extension, Up to 360% Faster MySQL. In addition, Linux 7.3, Intel Nova Lake S Integrated GPU to Receive Stable Support discusses Intel’s integrated GPU support in the same Linux 7.3 cycle.

The processors affected by this patch series are multiple generations of Intel processors. All processors adopting a hybrid architecture, from Alder Lake to Panther Lake, are covered. According to Neri’s explanation, the problem occurred on “several generations of Intel processors already shipped and upcoming.”

The improvement to cluster-distributed scheduling could affect both server workloads and desktop workflows that use hybrid CPUs on Linux. In particular, on systems where some cores are idle, the proper redistribution of tasks is expected to reduce performance variance.

Editorial Opinion

Short-term impact: The integration of this patch series is expected to bring direct benefits to Intel hybrid CPU users at the time of the Linux 7.3 release. On workstations and laptops equipped with Core (Ultra) series processors from Alder Lake onward, performance degradation caused by load-balancing imbalances may be reduced. In the server space, the kernel-side foundation will also be in place should Intel’s hybrid Xeon products be rolled out in the future. Long-term perspective: The fix to cluster scheduling can be seen as preparation of a technical foundation that will also extend to combinations with ARM big.LITTLE configurations and future asymmetric architectures. As asymmetric multiprocessing becomes mainstream on both x86 and ARM, how the kernel scheduler adapts to topology diversity is directly linked to the performance of future edge computing and mobile servers. Question from our editorial desk: This patch series is a fix specific to Intel topologies; whether improvements to cluster scheduling for ARM-based asymmetric systems are proceeding in parallel is a point that requires verification.

References

• “Linux 7.3 To Better Handle Cluster Load Balancing On Intel Hybrid CPUs”, by Michael Larabel — Phoronix, 2026-08-08T17:27:05.000Z (ARR)
• Source URL: https://www.phoronix.com/news/Linux-7.3-Better-Intel-Hybrid