iowait is the share of time the CPU sat idle while there was outstanding disk I/O. High iowait means storage is the bottleneck: run iostat -x 1 to find the device with high %util and await, then iotop or pidstat -d to find the process doing the reads or writes.
Symptoms
- top shows a high wa value in the %Cpu(s) line
- Load average is high while CPU usage is low
- Database queries and file operations slow down
How to diagnose it
Confirm iowait
The wa column shows the percentage. Sustained values above 10–20% deserve a look.
vmstat 1 5Find the busy device
%util near 100 means the device is saturated. r_await and w_await show average latency in milliseconds; tens of ms on an SSD is slow.
iostat -x 1 5 # from the sysstat packageFind the process
iotop -o shows only processes doing I/O. pidstat gives the same per process without root.
sudo iotop -o pidstat -d 1 5Check for disk errors
I/O errors, resets or timeouts in the kernel log suggest failing hardware.
sudo dmesg -T | grep -iE "i/o error|reset|timeout|blk_update"Check SMART health on physical disks
Look for reallocated or pending sectors and media errors.
sudo smartctl -a /dev/sda
Common causes and fixes
Database without enough memory for its working set
Increase the buffer pool or shared buffers if RAM allows, and add indexes for queries that scan tables.
Swapping
If si/so in vmstat are non-zero, the disk is busy with swap. Fix memory pressure first (see the swap guide).
Backups, log rotation or large copies
Run them at idle I/O priority and off-peak.
ionice -c3 nice -n 19 rsync -a /data /backupCloud volume burst credits or IOPS limit exhausted
On AWS gp2/burstable volumes or small cloud disks, check provider metrics for throttling and move to a volume with provisioned IOPS or throughput.
Failing disk or degraded RAID
Check SMART and RAID status (cat /proc/mdstat). Replace disks reporting errors before they fail completely.
Prevent it happening again
- Alert when CPU_iowait is above 20% for 10 minutes.
- Watch per-device utilisation and throughput history to spot a disk approaching its limits.
- Alert on SWAP_used so swapping does not silently turn into disk load.
Set it up: Linux monitoring · AWS EC2 monitoring.
FAQ
What is a good iowait percentage?
Close to zero on most servers. Short spikes during backups are normal; sustained double-digit iowait means storage is limiting performance.
Does high iowait mean the CPU is busy?
No. iowait is idle time during which I/O was pending. The CPU could do other work; processes are simply waiting for the disk.
Related guides
Get alerted before this becomes an outage.
Install the agent in one line. Statance watches CPU, memory, disk and services, and its AI assistant helps find the cause.