Analyzing I/O (iostat)

HolyCloud VPS performance often depends on disk I/O (local storage or NVMe). The iostat tool (sysstat package) shows throughput, utilization, and latency per device.

Prerequisites

• Linux VPS with sudo access
• Representative load (backup, SQL import, compile) to see peaks
• Storage type identification (HolyCloud panel / lsblk)

Installation

sudo apt install -y sysstat
sudo systemctl enable --now sysstat
iostat -V

On RHEL:

sudo dnf install -y sysstat

Real-time reading

Every 2 seconds, 5 samples:

iostat -xz 2 5

Important columns (extended mode -x):

| Column | Meaning |

|---------|-------------|

| %util | Time the disk was busy (100% = saturation) |

| r/s, w/s | Reads / writes per second |

| rkB/s, wkB/s | Read / write throughput |

| await | Average I/O latency (ms) — critical |

| avgqu-sz | Queue depth |

If %util is near 100% and await rises (e.g. > 20 ms on NVMe), the disk is the bottleneck.

Per partition

iostat -xz 2 5 /dev/vda
df -hT
lsblk -o NAME,SIZE,TYPE,MOUNTPOINT,FSTYPE

On a VPS, the disk often appears as vda, sda, or nvme0n1.

History with sar

sysstat stores archives:

sar -d -f /var/log/sysstat/sa$(date +%d) | tail -30

Overnight backup activity is visible the next day.

Typical scenarios

Slow MySQL import

iostat -xz 1
# in another terminal
mysql < gros_dump.sql

If w/s and wkB/s are high with high await, prefer off-peak imports or an appropriate innodb_buffer_pool_size.

rsync backup

watch -n2 'iostat -xz 1 1 | grep -E "Device|vda|nvme"'

Active swap

vmstat 1
free -h

If si/so (swap in/out) > 0 continuously, RAM is insufficient — iostat will also show disk reads.

Compare with fio (optional)

One-off test (warning: intensive load):

sudo apt install -y fio
fio --name=randread --ioengine=libaio --rw=randread --bs=4k \
  --numjobs=4 --size=256M --runtime=30 --filename=/tmp/fiotest
rm -f /tmp/fiotest

Compare results between two VPS Performance plans.

Best practices

• Measure during an incident, not only at idle.
• Note the timezone and parallel cron job (grep CRON /var/log/sysstat).
• Do not confuse high CPU and I/O: use top, mpstat as well.

apt install -y sysstat
mpstat -P ALL 1 3

Indicative thresholds (NVMe VPS)

| Metric | Comfortable zone | Alert |

|----------|------------------|--------|

| await | < 2 ms | > 10 ms sustained |

| %util | < 70% average | 100% prolonged |

| avgqu-sz | < 2 | > 10 |

Thresholds vary by workload; an OLTP database server tolerates less latency than a static site.

Troubleshooting

| Observation | Approach |

|-------------|-------|

| Low %util but slow app | Network, CPU, application locks |

| Huge w/s, little CPU | Logs, tmp, antivirus scan |

| Constant reads | Cold cache, missing SQL indexes |

Need help?

Provide HolyCloud support: iostat -xz 2 10 output during slowness, df -h, VPS plan, and exact incident time.