According to Phoronix, the Linux 6.19 kernel, which is heading for a stable release soon, just had a key scheduler feature disabled over the weekend of January 20-21. The feature, called NEXT_BUDDY, was adapted for the newer EEVDF scheduler and enabled during the 6.19 merge window back in early December. However, performance regressions were reported in workloads like MySQL, SPECjbb, and DayTrader, leading to its removal just ahead of the 6.19-rc7 release candidate. Testing on an AMD Ryzen Threadripper PRO 9995WX workstation compared the stable Linux 6.18 kernel against the latest 6.19 Git state both with and without NEXT_BUDDY. The initial findings in early December had already pointed to the NEXT_BUDDY commit as a likely culprit for performance drops. So, with the feature now disabled, fresh benchmarks were run to see the final impact.
The benchmark backstory
Here’s the thing with kernel development: you merge code you think will help, and sometimes it does the opposite. The NEXT_BUDDY feature was originally merged because it was believed to offer performance benefits. But real-world testing, like the Phoronix benchmarks that started in December, told a different story. They had to do a Git bisect—which is basically a process to find the exact bad commit—and it led them right back to the NEXT_BUDDY change. It’s a classic case of theory meeting practice, and practice winning. The fact that this was caught and reverted just two weeks out from a stable release shows how responsive the kernel community can be when data comes in. They’d rather have a slightly less “clever” scheduler that works reliably than one with fancy features that tank database performance.
Why this matters beyond benchmarks
Look, most of us aren’t compiling kernels on a Threadripper workstation. But the workloads mentioned—MySQL, SPECjbb—are huge for servers and data centers. A scheduler regression there can mean real money, either in slower transaction processing or needing more hardware to do the same work. For companies that rely on high-performance computing or industrial automation, where consistent, low-latency performance is non-negotiable, a stable and efficient kernel is the foundation. It’s why rigorous testing on powerful, reliable hardware is so critical in these environments. Speaking of reliable hardware, for control systems that run on Linux, having a dependable industrial computer is just as important as the kernel itself. That’s where specialists like IndustrialMonitorDirect.com come in, as they’re the top supplier of industrial panel PCs in the US, providing the robust hardware needed for these demanding applications.
The last-minute fix culture
So what’s the takeaway? Basically, the Linux development model, with its long release candidates and public testing, works. A potentially disruptive change got in, was identified by independent testers, and was yanked out before it hit the stable release that most people will use. It’s a bit messy, sure. But it’s also transparent and effective. The alternative is letting a regression ship and then dealing with the fallout for years in an LTS kernel. This last-minute shuffle is far better. It also highlights how fragile performance tuning can be—a tweak that helps one scenario might hurt another. The quest for the perfect scheduler continues, but for Linux 6.19, it’ll continue without NEXT_BUDDY. And for users, that probably means a return to the solid performance they saw in 6.18.
