Linux 6.19 Slated To Land "mm/cid" Rewrite That Has Very Positive Performance Potential
A set of Linux kernel patches posted back in October for rewriting the kernel's memory-mapped concurrency ID code for some nice performance wins looks like it will land for Linux 6.19. This is the code that prominent Intel engineer Thomas Gleixner found to yield up to an 18% improvement for the PostgreSQL database. My testing of this "mm/cid" code has also shown some nice performance wins too.
Intel Fellow Thomas Gleixner overhauled the CID management code after finding the existing complex code introduced significant overhead into the kernel scheduler's hot code paths. This new code is simpler and lower-overhead.
Queued today into tip/tip.git's core/rseq Git branch is this rewrite of the memory management's concurrency ID (CID) code. With the patches from Gleixner now being in a TIP branch, they will likely be submitted for the upcoming Linux 6.19 merge window barring no last minute problems or objections raised by prominent developers like Linus Torvalds himself.
This includes reverting the complex CID management:
And then from there proceeding to overhaul the mm/cid code with the new implementation.
A few weeks back I ran some benchmarks of the tentative code within Intel's Rewrite Of Linux MM CID Code Showing Some Nice Gains For AMD.
Linux 6.19 continues looking more exciting by the day. The Linux 6.19 merge window will happen in early December while the stable kernel should be out in February.
Intel Fellow Thomas Gleixner overhauled the CID management code after finding the existing complex code introduced significant overhead into the kernel scheduler's hot code paths. This new code is simpler and lower-overhead.
👁 mm/cid patches queued in TIP branch
Queued today into tip/tip.git's core/rseq Git branch is this rewrite of the memory management's concurrency ID (CID) code. With the patches from Gleixner now being in a TIP branch, they will likely be submitted for the upcoming Linux 6.19 merge window barring no last minute problems or objections raised by prominent developers like Linus Torvalds himself.
This includes reverting the complex CID management:
"The CID management is a complex beast, which affects both scheduling and task migration. The compaction mechanism forces random tasks of a process into task work on exit to user space causing latency spikes.
Revert back to the initial simple bitmap allocating mechanics, which are known to have scalability issues as that allows to gradually build up a replacement functionality in a reviewable way."
And then from there proceeding to overhaul the mm/cid code with the new implementation.
A few weeks back I ran some benchmarks of the tentative code within Intel's Rewrite Of Linux MM CID Code Showing Some Nice Gains For AMD.
Linux 6.19 continues looking more exciting by the day. The Linux 6.19 merge window will happen in early December while the stable kernel should be out in February.