[ad_1]
Very large IBM mainframes / servers took over 30 minutes to boot the Linux kernel … No, not for system POST with memory learning etc., but for Linux loading. Fortunately, with the Linux 5.15 kernel there is a set of scalability improvements to get these large IBM systems up and running in about five minutes.
Along with the basic driver changes for Linux 5.15, there is a set of fixes that improve Kernfs performance for features used around pseudo-filesystems like sysfs. Engineers discovered that large IBM Power mainframe systems with “several hundred processors and 64TB of RAM” took more than 30 minutes to boot the Linux kernel. Additional kernel parameters were also needed to prevent the kernel from timeing out on startup.
Extremely long boot times for modern high-end servers turn out to be the result of extensive path searches for non-existent files and extreme locking contention in VFS code.
Worse, with the 64TB of system memory and IBM Power splitting it into local 256MB blocks exposed through sysfs, a large number of sysfs nodes are created.
With the Kernfs scalability enhancements to sysfs found in the Base Driver Changes for Linux 5.15, these IBM systems can go from over 30 minutes booting to under five minutes now. The changes involve changing the Kernfs mutex to using a read-write semaphore to allow parallel node lookups, improved path resolution, and the use of VFS negative dentry caching. .
These Kernfs enhancements and many more can be found through the main driver PR that was merged today for Merge Window 5.15.
[ad_2]
