AIX Performance Management

IBM AIX is engineered for reliability and throughput, but no system is immune to performance degradation. Performance problems rarely announce themselves as “CPU issues” or “disk issues”—they surface as slow applications, timeouts, paging storms, or missed SLAs.

Effective AIX performance management is systemic, not reactive. It follows a continuous lifecycle:

Monitor → Analyze → Tune → Validate → Report → Repeat

This guide covers what to measure, why it matters, how to tune it, and how to avoid common traps—with real AIX commands and proven thresholds.

The Performance Management Lifecycle

• Monitor: Track CPU, memory, disk I/O, and network in real-time using tools like topas, nmon, and vmstat.
• Analyze: Spot bottlenecks through trends with sar and nmon reports.
• Tune: Adjust kernel parameters (vmo, ioo, no), apps, or hardware.
• Report: Summarize data for audits using nmon and custom scripts.

Pro Tip: Generate HTML reports effortlessly with nmon in analyzer mode:

# nmon -f perf.nmon

# nmonanalyser perf.nmon

CPU Performance: 

CPU metrics quickly reveal if user processes or the kernel are hogging cycles.

Key Metrics

Metric	Meaning	Healthy Range
%user	User processes	<70%
%sys	Kernel time	<30%
%idle	Idle CPU	>20%
%wait	I/O wait	<10%
Load Avg	Processes waiting (1/5/15m)	< #cores

Bottleneck Signs

• High %user: CPU-bound apps (e.g., heavy compiles).
• High %sys: Kernel thrashing (network or filesystem issues).
• Low %idle + high load: Add cores or tune processes.
• High %wait: Address I/O bottlenecks first.

Top Tools & Commands

# vmstat 5        # CPU stats every 5s

  kthr   -----memory---------- ---swap-- -----io--- ---CPU---

  r  b   avm  fre  fi fo pi po fr  sr  in  sy  cs us sy id wa

  1  0  1024 512   0  0  0  0 10  5 100 200  50 20 10 60 10

# mpstat 1 5      # Per-CPU stats

CPU   %usr %sys %wio %idle

 0    15    5    2    78

 1    25   10    5    60

Other essentials: topas (interactive; 't' for top processes), nmon ('c' for CPU, 'C' for per-core), sar -u 1 10 (historical data).

Tune It: Check scheduler with schedo -i; bind apps to cores via bindprocessor.

Memory Performance: Avoid Paging Hell

Memory shortages trigger paging and swapping, which spike CPU %wait and kill performance.

Key Metrics

• Free RAM: vmstat -v or svmon -G.
• Paging: pgspcn (pages scanned), pgin/pgout.
• Swap: lsps -a; keep usage <20%.

Bottleneck Signs

• High paging + low free RAM: Time to buy more memory.
• High cache but thrashing: Optimize inefficient apps.

Commands

# svmon -G        # Global summary

Memory Size:          16384 MB

Fixed:                1024 MB

Persistent:            256 MB

Client:             15000 MB

Pinned:                512 MB

Free:                  800 MB   # Low? Alert!

Also try svmon -P (top processes), vmstat 5 (paging columns like pi/po/fr/sr), and topas (memory view).

Tune It:

# vmo -a          # Check current params

vmo -p -o minperm%=3     # Boost file cache

vmo -p -o maxperm%=80    # Cap computational memory

vmo -p -o maxclient%=80  # Limit client memory

Disk I/O: Speed Up Your Storage

Slow disks throttle apps—focus on IOPS, queues, and service times.

Key Metrics

Metric	Meaning	Target
%busy	Disk utilization	<70%
Queue Length	Waiting requests	<2
Service Time	Avg response (ms)	<20ms
IOPS	Ops/sec	Device max

Bottleneck Signs

• High %busy or queue: Overloaded hdisk.
• High CPU %wait: Disk-bound workload.

Commands

# iostat -d 5 3   # Disk I/O stats

     hdisk0  hdisk1

KB/t  tps KB/t  tps

32.5 150 45.2 200   # High tps? Rebalance

More: lsvg -p rootvg (VG to hdisk map), lsattr -El hdisk0 (attributes like queue_depth), topas ('d' for disk detail), nmon ('D' disks, 'j' JFS).

Tune It:

• Stripe LVs: mklv -S stripe rootvg 4 hdisk1 hdisk2 hdisk3 hdisk4.
• Use JFS2: mkfs -V jfs2 /dev/lv.
• Boost queue: chdev -l hdisk0 -a queue_depth=32.

Network Performance: Keep Data Flowing

Network issues hammer NFS, databases, and clusters.

Key Metrics

• Throughput: pkt/s, B/s.
• Errors: drops, collisions, retransmits.
• Latency: ping or app response times.

Bottleneck Signs

• High errors: Faulty cable or switch.
• Low throughput: Undersized buffers.

Commands

# netstat -i      # Interface stats

Input     Output

ent0:   1.2G     800M   # Packets/errors

Try entstat -d ent0 (detailed), netstat -an (TCP/UDP), topas ('n' network), nmon ('N' nets), tcpdump -i ent0 (capture).

Tune It:

# no -p -o tcp_sendspace=262144  # Larger buffers

# no -a | grep tcp               # Verify

# chdev -l ent0 -a mtu=9000      # Jumbo frames

Quick-Start Dashboard Script

Drop this into a ksh script for instant AIX perf snapshots:

#!/bin/ksh

echo "AIX Perf Snapshot $(date)"

vmstat 1 3 | tail -1

svmon -G | head -5

iostat -d 1 2 | tail -1

netstat -i | tail -1

Bookmark this guide, run these commands, and watch your AIX systems fly. Got a specific bottleneck? Dive in and tune!