How to identify bottlenecks on Linux

To identify Linux system bottlenecks, you should first check the following four aspects. 1. The CPU usage is too high: Use top, htop or mpstat to check to find out the processes with high usage and optimize or limit their resource usage. 2. Insufficient memory leads to frequent swapping: check Swap usage through free -h, /proc/meminfo or vmstat, increase memory, reduce memory consuming programs or adjust the swappiness parameter. 3. Disk I/O bottleneck: Use iostat -x to monitor disk utilization and response time, optimize applications, upgrade storage media, or enable caching mechanisms. 4. Network delay or bandwidth is full: analyze traffic through iftop, nload or sar -n DEV, optimize transmission, limit abnormal access or use CDN. During the investigation, each case should be analyzed one by one to avoid misjudgment.

Identifying bottlenecks in a Linux system is not difficult, the key is knowing where to start. Slow system, delayed response, and high resource utilization may all be symptoms of bottlenecks. But to really find the problem, you need to investigate in an organized manner.

The following areas are common sources of bottlenecks and are areas you should check first.

1. CPU usage is too high

The CPU is one of the components most prone to bottlenecks. If you find that the system responds slowly, especially when performing tasks, it may be that the CPU is insufficient.

How to check:

• Use top or htop to view overall CPU usage.
• Use mpstat (from the sysstat package) to view the load on each CPU core.
• If you see that %sy (system call) is very high, it means that the kernel is busy; if %us (user mode) is very high, it may be that a process is occupying the CPU.

what to do:

• Find the process with the highest CPU usage: ps aux --sort=-%cpu | head
• Check whether the program logic can be optimized or its CPU usage limited (such as using cpulimit )
• If it is frequently full, consider upgrading the hardware or doing load balancing

2. Insufficient memory leads to frequent swapping (Swap)

When there is not enough memory, Linux will use Swap space to relieve pressure, but this will cause a significant drop in performance. Swap is used more often, which usually means that memory has become a bottleneck.

How to check:

• Look at the output of free -h , especially the usage of Swap
• Observe /proc/meminfo or use vmstat
• If si/so (swap in/out) continues to be greater than 0, it means frequent swapping

what to do:

• Increase physical memory (the most direct and effective)
• Reduce the number of memory-intensive programs
• Adjust swappiness parameter ( /proc/sys/vm/swappiness ) to control swapping tendency
• Check for memory leaks (such as Java applications)

3. Disk I/O becomes a bottleneck

Disk read and write speeds cannot keep up with requests, which can cause the entire system to slow down, especially when processing a large number of logs, database operations, or virtual machines.

How to check:

• Use iostat -x to view %util and await of the device
• If %util is close to 100%, the disk is saturated
• A high await (for example, more than 10ms) indicates a slow response

what to do:

• Optimize applications to reduce unnecessary IO
• Use faster storage media (such as SSD instead of HDD)
• Spread data across multiple disks (RAID or LVM)
• Enable caching mechanisms (such as Redis, Memcached)

4. Network delay or bandwidth is full

Network bottlenecks usually occur when communicating between servers, uploading and downloading large files, or being attacked by DDoS.

How to check:

• Use iftop or nload to view real-time bandwidth usage
• sar -n DEV can see historical traffic trends
• If you find that the bandwidth is full or packet loss is severe, it means there is a problem with the network.

what to do:

• Upgrade bandwidth or optimize transmission protocols (such as compression, chunking)
• Analyze traffic sources to prevent abnormal access (such as iptables speed limit)
• Use CDN to ease public network access pressure
• For LAN issues, check the switch or network card configuration

The bottleneck is often not a single one, but is often caused by the superposition of multiple factors. Therefore, you must be in order when investigating, one by one, and don't rush to conclusions.

That's basically it. Starting from these four aspects, most common bottlenecks can be located.

The above is the detailed content of How to identify bottlenecks on Linux. For more information, please follow other related articles on the PHP Chinese website!