Kernel developers have merged a series of scheduler updates for Linux 7.1. These changes target efficiency in multi-core systems. No sweeping overhaul here. Instead, targeted fixes aim to boost performance where it counts.
Peter Zijlstra, Intel’s veteran scheduler expert, led one standout patch. His more complex proportional newidle balance code promises big wins. Workloads like the easyWave simulation benchmark and FIO storage tests “benefit greatly,” according to testing notes from Phoronix. Zijlstra’s patch, detailed in the kernel mailing list, refines how the scheduler balances tasks on newly idle CPUs. It weighs load distribution more precisely across cores.
Vincent Guittot contributed another key piece. His update to over-utilized detection sharpens how the kernel spots overloaded CPUs. That matters on busy servers. AMD’s K Prateek Nayak optimized topology code too. His series cuts allocation overhead in scheduler domains, aiding AMD’s multi-chiplet designs.
The full pull request, submitted by Tejun Heo, lists dozens of these incremental patches. They build on the EEVDF scheduler, now the default fair-policy engine since Linux 6.12 fully retired CFS. EEVDF, or Earliest Eligible Virtual Deadline First, picks tasks by virtual deadlines. It favors low-latency ones without starving others. The kernel docs explain: “EEVDF aims to distribute CPU time equally among all runnable tasks with the same priority,” as outlined in kernel.org documentation.
EEVDF’s Foundation Shapes 7.1’s Incremental Push
EEVDF arrived amid controversy. Peter Zijlstra proposed it in 2023 to fix CFS’s accumulated hacks. CFS, born in 2007, piled on heuristics for sleepers and latency. EEVDF cleans that up. A 1995 paper by Ion Stoica and Hussein Abdel-Wahab inspired it. Linux phased it in: optional in 6.6, sole fair scheduler by 6.12.
But real-world quirks persist. A Reddit thread from embedded developers notes race conditions on single-core setups. “EEVDF is more aggressive about preemption than CFS under equal niceness,” one poster explained on r/embedded. Code with timing assumptions broke. That’s exposed latent bugs.
Sched_ext evolves alongside. Linux 7.1 lays cgroup sub-scheduler groundwork. BPF programs can now attach hierarchically to cgroups. That lets admins mix policies for diverse loads—gaming here, batch jobs there. Phoronix covered the merge: groundwork for “multiple BPF schedulers… dispatch path is made hierarchical,” per the pull request. Idle SMT sibling selection got tweaks too. Documentation and latency controls improved.
HRTIMER overhaul ties in. Thomas Gleixner’s rework cuts overhead for frequent timers like HRTICK. “A hrtick enabled scheduler provides the same performance as without hrtick,” Gleixner noted in his Linux 7.1 pull. HRTICK boosts responsiveness. Now it’s cheaper.
Benchmarks loom. Phoronix plans Linux 7.1 tests next week. Early signs point to gains in I/O-heavy and simulation tasks. Servers might see steadier balance on AMD EPYC or Intel Xeon. But watch 32-bit x86. VFS changes could drag there, per separate Phoronix reports.
Industry watchers note broader context. Sched QoS prototypes hint at user-space hints, echoing Apple’s classes. Qais Yousef’s alpha, inspired partly by iOS, aims for finer control. Not in 7.1 yet. Cache-aware scheduling missed the window too.
X chatter echoes the buzz. Phoronix tweeted the scheduler news on April 18, drawing views from kernel fans. CachyOS rolled out related user-space tools for low-VRAM setups. Embedded folks gripe about preemption shifts.
These 7.1 changes refine a maturing system. EEVDF handles the core. Sched_ext opens extensibility. Timers and balance polish edges. Expect Phoronix benchmarks to quantify. Servers, desktops, embedded—all stand to gain. Or regress, in spots. Testing will tell.
WebProNews is an iEntry Publication