Docs

Elastic Compute Service

Memory issue

Why does prolonged high I/O or unresponsiveness occur before OOM in veLinux 1.0 CentOS Compatible Edition image compared to a community-version CentOS image?

Causes

The kernel version of the veLinux 1.0 CentOS Compatible Edition image is 5.x, which is higher than that of the community CentOS image. Kernels of higher versions have more aggressive memory reclamation policies.

In scenarios involving high performance, availability, and throughput (for example, high-concurrency web servers, big data processing, and virtualized environments), a more aggressive memory reclamation policy can help you utilize memory, reduce memory fragmentation, and avoid OOM events more effectively. However, when memory usage reaches the limit, the kernel of a higher version typically spends more time reclaiming memory. At the same time, binary files of system services are frequently swapped in and out, causing increased system I/O load. This may cause degraded system performance or even an unresponsive system.

Solutions

Given the mentioned reasons, you can set a memory upper limit for services to prevent memory usage from reaching the limit, or use a CentOS image with a lower kernel version to avoid this issue.

Reference solutions are as follows:

Limit the memory of a process

Execute the following command to set the memory limit for the target service.

tip

Replace xx.service with the target service.

Replace 100M with the memory limit you need. Unit: MB.

systemctl set-property xxx.service MemoryLimit=100M

Execute the following command to confirm whether the memory limit of the target service is successfully modified.

tip

Replace xxx.service with the target service.

systemctl status xxx.service

Use a CentOS image with kernel version 3.x. For more information, see Changing operating systems.

Last updated: 2026.04.13 16:50:03