High load average: is it CPU, disk or memory?
A high load average does not say what is slow. Use uptime and vmstat to tell CPU saturation from blocked I/O or swapping, then find the process.
What you see
uptimeshows load averages well above the number of CPU cores- The server feels slow, but
topmay show low CPU use
What it means
The three load numbers are averages over 1, 5 and 15 minutes of processes that are either runnable or in an uninterruptible state, which usually means waiting on I/O. So load counts CPU demand and disk waits together.
That is why high load with idle CPU is possible. Split the two first, because the fix for each is different.
How to confirm the cause
- 1Read the three load numbers and compare them to your core count
uptime && nproc
A 1-minute figure above the 15-minute one means load is rising now. Load near the core count is full use, not necessarily trouble.
- 2Split CPU demand from I/O waiting
vmstat 1 5
ris runnable processes,bis processes blocked waiting for I/O,wais time spent waiting for I/O,siandsoare memory swapped in and out per second. Ignore the first line; it is the average since boot. - 3Find the process responsible
top
Press
Pto sort by CPU andMto sort by memory. Processes in stateDare the ones waiting on I/O.
Causes and fixes
CPU saturation
- HOW TO TELL
ris consistently above the core count andwais low.- ROOT CAUSE
- More processes want to run than there are cores, so they queue. It is usually one busy process or a traffic spike.
- FIX
- Find the busy process with
top, then fix, throttle or scale it. If demand is real, add cores or instances.
Blocked on disk or network storage
- HOW TO TELL
bis high andwais high while CPU use is low.- ROOT CAUSE
- Processes are stuck in uninterruptible I/O waits, so they count toward load without using CPU. A slow or failing disk, or a hung network mount, is the usual source.
- FIX
- Check the disk and any network filesystems. Look for I/O errors in the kernel log with
journalctl -k -b.
Memory pressure and swapping
- HOW TO TELL
siandsoare non-zero for long stretches.- ROOT CAUSE
- When RAM runs out the kernel swaps pages to disk, and every swap is slow I/O. The system spends its time moving memory instead of working.
- FIX
- See the heavy-swapping runbook. Reduce memory use or add RAM.
How to stop it coming back
- Alert on load per core and on
wa, not on load alone. - Record
vmstatoutput during incidents so the cause is on file. - Know your normal load so a spike stands out.
Keep going
- Runbook: Server is crawling: heavy swapping
- Runbook: Process killed by the out-of-memory killer
- Cheat sheet: Linux
- Cheat sheet: journalctl
- Lesson: the linux topic in Learn
Sources
Linux man-pages (man7.org) and the systemd manual (freedesktop.org), read 18 Nov 2026. Command options were checked against those pages.