Suboptimal performance due to long IBM Spectrum Scale waiters
Low-level system issues, like slow disks, or slow network, might cause long GPFS™ waiters. These long waiters cause performance degradation. You can use the mmdiag --waiters command to display the mmfsd threads waiting for events. This information can help resolve deadlocks and improve the system performance.
Problem identification
On the node, issue the mmdiag --waiters command to check whether any long waiters are present. The following example shows long waiters that are contributed by the slow disk, dm-14:
#mmdiag --waiters0x7FF074003530 waiting 25.103752000 seconds, WritebehindWorkerThread: for I/O completion on disk dm-14
0x7FF088002580 waiting 30.025134000 seconds, WritebehindWorkerThread: for I/O completion on disk dm-14Problem resolution and verification
Resolve any system-level or software issues that exist. When you verify that no system or software issues are present, issue the #mmdiag --waiters command again to verify whether any long waiters exist.
One possible reason for long waiters, among many, can be that
Samba lock directory has been configured to be located in GPFS.