High Linux load but low CPU usage. Where would you start?

November 19, 2026 5 answers
Asked by 7a!claspgizmo_835

Application responses are slowing down, load average keeps rising, and CPU usage is around 20 to 25 percent. Several processes are in D state, and the application uses an NFS mount.

How would you figure out what the processes are waiting for and distinguish an NFS problem from local disk trouble?

Answers

5k+claspnotch_827

Start with vmstat 1 and watch the runnable and blocked process counts alongside CPU and swap activity. On Linux, load average includes tasks waiting in uninterruptible sleep as well as runnable tasks, so high load does not necessarily mean busy CPUs.

Use ps or top to find the D-state tasks, then inspect /proc/PID/stack for a few of them if you have permission. The kernel stack can show where a task is parked. NFS or RPC functions are a useful clue, but a single function name is not always enough to prove the root cause.

7h~gimbalcrank_429:

Kernel stack access may require elevated privileges, and the function names can vary by kernel. Look for a pattern across several stuck tasks.

Reply to this answer
v.739+ferrulepiston

Check the NFS client before assuming the mount itself is broken. nfsstat -rc can show client RPC retransmissions, and nfsiostat can show per-operation latency and retransmissions for mounts. Compare counters over time rather than treating a single total as proof of a current fault.

Then check network connectivity and errors between the client and server, plus the NFS server's own load and storage health if you have access. A slow server, network loss, or a server-side storage problem can all leave the client waiting.

p.937~tinselcrank:

Commands that touch the mount, including df, can themselves hang when the server is unresponsive. Use care when running them during an incident.

Reply to this answer
Mavdrix+8h

For local disks, run iostat -x 1 and look at latency and queue depth over several samples. Check kernel logs for disk errors, timeouts, or bus resets too. A busy device alone does not tell you whether it is the bottleneck.

Compare that with the NFS stats and the blocked task stacks. iotop can help identify local I/O activity, but it is not a reliable way to rule out NFS waits.

Reply to this answer
v.625~zordavi

Don't forget memory pressure. Check swap activity in vmstat and available memory, because heavy swapping can make the whole system crawl. If this is a virtual machine, check the host or hypervisor metrics too. The guest may not show the full picture of contention for physical resources.

Reply to this answer
r.468+bramblewicket

I wouldn't start by unmounting the share or rebooting a production server. The application may depend on that mount, and either action can disrupt it or erase useful evidence. Gather the task stacks and storage metrics first, then decide on recovery steps with the service impact in mind.

Reply to this answer

Add your answer

Answers are reviewed before publication.

Related Discussions