Linux memory usage experience

Linux system memory usage experience-general Linux technology-Linux programming and kernel information. The following is a detailed description. I read many articles about memory leak and linux memory management, which are caused by misunderstandings due to lack of linux memory management. In the evening, I will summarize some of my thoughts and experiments I 've seen over the past few days for your reference. If you have any mistakes, please point them out directly.

Linux memory management, in fact, is very similar to windows Memory Management. It uses the concept of virtual memory, so I have to scold MS here, why do we still use pagefile when there is a lot of physical memory. this is why we often have to argue with a group of people about the size of Pagefile and how to allocate it. in Linux, we don't have to argue about the size of swap. I personally think, it is enough to set up a m swap. What if the m swap is not enough? You can only say that you still need to add memory, or check whether your application actually has a memory leak.

In Linux, we usually use command free to view memory.

[Root @ nonamelinux ~] # Free

Total used free shared buffers cached

Mem: 386024 377116 8908 0 21280 155468

-/+ Buffers/cache: 200368 185656

Swap: 393552 0 393552

The following is an explanation of these values:

Row 2 (mem ):

Total: total physical memory size.

Used: used.

Free: available.

Shared: The total memory Shared by multiple processes.

Buffers/cached: disk cache size.

Row 3 (-/+ buffers/cached ):

Used: used.

Free: available.

The fourth line won't be explained much.

Differences:

The difference between used/free of the second row (mem) and used/free of the third row (-/+ buffers/cache.

The difference between the two is that from the usage perspective, the second line is from the OS perspective, because for OS, buffers/cached is used, so its available memory is 8908KB, the used memory is kb, including the + buffers + cached used by the kernel (OS) + Application (X, oracle, etc.

The third line indicates that, from the application perspective, for applications, buffers/cached is equivalent to available, because buffer/cached is designed to improve file read performance, when the application needs to use the memory again, the buffer/cached will be quickly recycled.

From the application perspective, the available memory = system free memory + buffers + cached.

For example:
185656 = 8908 + 21280 + 155468

Next we will explain when the memory will be exchanged and how it will be exchanged.

When the available memory is less than the rated value, the switch starts.

Rating (RHEL4.0 ):

# Cat/proc/meminfo

Switching reduces the number of physical pages used in the system in three ways:

1. Reduce the buffer and page cache size,

2. swap out the V-type memory page of the system,

3. Swap out or discard the page. (The Memory Page occupied by the Application, that is, the physical memory is insufficient ).

In fact, using swap in a small amount affects the system performance.

The following is the difference between buffers and cached.

Buffers is the buffer size for Block devices. It only records the metadata and tracking in-flight pages of the file system.

Cached is used to buffer files.

That is to say: buffers is used for storage. What content and permissions are contained in the directory.

Cached is used directly to remember the files we opened. If you want to know whether the files take effect, you can try and execute two commands one by one # man X, you can obviously feel that the second start is much faster.

Experiment: it is obvious on a machine with no application. Remember that the experiment can only be performed once. If you want to do more, change the file name.

# Free

# Man X

# Free

# Man X

# Free

You can compare the size of buffers after free.

Another experiment:

# Free

# Ls/dev

# Free

You can compare the two sizes. Of course, the buffers are increasing at any time, but if you have ls, the increase will be faster. This is the difference between buffers and chached.