Master Linux Performance: Key Factors and Essential Optimization Tools

1. Factors Affecting Linux Performance

1. System Hardware Resources

(1) CPU

How to determine multi‑core CPUs versus hyper‑threading; CPU‑intensive services include dynamic web services and mail services.

(2) Memory

Trade‑off between physical memory and swap.

Choosing a 64‑bit Linux distribution.

Memory‑intensive services include in‑memory databases such as Redis, HBase, and MongoDB.

(3) Disk I/O

RAID technologies (RAID0/1/5/01/10).

SSD disks.

Disk‑intensive workloads are typically database servers.

(4) Network Bandwidth

Selection of NICs and switches.

Operating‑system level NIC bonding.

Bandwidth‑heavy services include Hadoop platforms and video streaming platforms.

2. OS‑Related Resources

(1) System Installation Optimization

Disk partitioning, RAID configuration, and swap settings.

(2) Kernel Parameter Tuning

ulimit -n (maximum open files) and ulimit -u (maximum user processes).

(3) File System Optimization

ext2 – standard Linux file system without journaling.

ext3 – adds journaling; supports up to 32,000 sub‑directories.

ext4 – successor to ext3, unlimited sub‑directories, fast fsck (supported since kernel 2.6.28).

xfs – high‑performance file system, default since kernel 3.10.

Recommendations:

Read‑intensive workloads with many small files: prefer ext4, then xfs, then ext3.

Write‑intensive workloads: prefer xfs, then ext4, then ext3.

Low performance or low data‑safety requirements: ext3 is a good choice.

3. Application Issues

These require developers to inspect code, but operations staff should provide strong evidence of program‑related problems.

2. Linux Performance Optimization Tools

1. CPU Performance Evaluation Tools

(1) vmstat (built‑in)

Monitors memory, process states, and CPU activity. Common usage: vmstat 2 3 (updates every 3 seconds, three samples).

Key fields:

procs – r column: number of processes running or waiting for a CPU slice; sustained values higher than CPU count indicate CPU shortage.

procs – b column: processes waiting for I/O or swapping.

memory – swpd: memory swapped out (KB). Non‑zero values are normal if si/so remain zero.

memory – free: free physical memory (KB).

memory – buff: buffer cache size.

memory – cache: page cache size; large cache with low bi indicates efficient filesystem.

swap – si: memory swapped in from disk.

swap – so: memory swapped out to disk; sustained non‑zero values indicate memory pressure.

IO – bi: blocks read per second; bo: blocks written per second. bi+bo > 1000 with high wa suggests disk I/O problems.

system – in: interrupts per second; cs: context switches per second. High values increase kernel CPU usage.

CPU – us: user‑space CPU usage; sy: kernel‑space CPU usage. us + sy > 80 % may indicate CPU saturation.

CPU – id: idle percentage.

CPU – wa: I/O wait percentage; wa > 20 % signals serious I/O wait.

(2) iostat (requires sysstat package)

Provides I/O statistics. Common usage: iostat -c 3 5 (CPU stats every 3 seconds, five samples). Use -d for disk stats.

(3) uptime

Shows current time, system uptime, number of logged‑in users, and load averages for 1, 5, 15 minutes.

2. Memory Performance Evaluation

(1) free

Shows memory usage. Typical usage: free -m. Important columns are free and cached. Example analysis: a system with 8 GB RAM, 925 MB free, 243 MB buffer cache, 6299 MB page cache, leaving ~7468 MB usable memory, indicating ample memory resources.

Rule of thumb: usable memory / total physical memory > 70 % → resources are abundant; <20 % → memory shortage; 20‑70 % → generally acceptable.

(2) sar / pidstat

Monitor CPU, memory, and I/O for all or specific processes. Common options: -u (CPU), -r (memory), -d (disk). Example commands: sar -u 3 – CPU stats every 3 seconds. pidstat -r -p 1 3 – memory stats for PID 1 over 3 seconds.

Output fields include Kbmemfree, Kbmemused, %memused, kbbuffers, kbcached, kbcommit, and %commit. sar’s output aligns with free but adds percentages and averages.

3. Disk Performance Evaluation

(1) iostat -d

Example: iostat -d 2 3. Output fields: Blk_read/s – blocks read per second. Blk_wrtn/s – blocks written per second. Blk_read – total blocks read. Blk_wrtn – total blocks written.

(2) pidstat -d -p <pid> 3

(3) sar -d 2 3

Provides detailed disk I/O metrics such as tps, rd_sec/s, wr_sec/s, avgrq‑sz, avgqu‑sz, await, svctm, and %util. Interpretation: avgrq‑sz – average request size (like items per shopper in a supermarket). avgqu‑sz – average queue length (people waiting). await – average wait time per I/O request. svctm – average service time (cashier speed). %util – proportion of time the device was busy; values near 100 % indicate a bottleneck.

Generally, svctm should be less than await. A high await relative to svctm signals long I/O queues and degraded performance.

4. Network Performance Evaluation

(1) ping

Shows round‑trip time (ms) and packet loss; high latency or loss indicates network issues.

(2) netstat

Use netstat -i to view interface statistics and netstat -r for routing tables.

(3) mtr / traceroute

Trace network routes; mtr provides dynamic, real‑time path analysis, useful for diagnosing network problems.

3. System Performance Analysis Standards

These standards combine the metrics described above to form a comprehensive view of CPU, memory, disk, and network health, guiding administrators in capacity planning and troubleshooting.