In this article, by Jonathan Hobson, the author of Troubleshooting CentOS, we will learn about memory management, swap, and swappiness.
(For more resources related to this topic, see here.)
A deeper understanding of the underlying active processes in CentOS 7 is an essential skill for any troubleshooter. From high load averages to slow response times, system overloads to dead and dying processes, there comes a time when every server may start to feel sluggish, act impoverished, or fail to respond, and as a consequence, it will require your immediate attention.
Regardless of how you look at it, the question of memory usage remains critical to the life cycle of a system, and whether you are maintaining system health or troubleshooting a particular service or application, you will always need to remember that the use of memory is a critical resource to your system. For this reason, we will begin by calling the free command in the following way:
# free -m
The main elements of the preceding command will look similar to this:
Total used free shared buffers cached
Mem: 1837 274 1563 8 0 108
-/+ buffers/cache: 164 1673
Swap: 2063 0 2063
In the example shown, I have used the -m option to ensure that the output is formatted in megabytes. This makes it easier to read, but for the sake of troubleshooting, rather than trying to understand every numeric value shown, let’s reduce the scope of the original output to highlight the relevant area of concern:
-/+ buffers/cache: 164 1673
The importance of this line is based on the fact that it accounts for the associated buffers and caches to illustrate what memory is currently being used and what is held in reserve. Where the first value indicates how much memory is being used, the second value tells us how much memory is available to our applications. In the example shown, this instance translates into 164 MB of used memory and 1673 MB of available memory.
Bearing this in mind, let me draw your attention to the final line in order that we can examine the importance of swap:
Swap: 2063 0 2063
Swapping typically occurs when memory usage is impacting performance. As we can see from the preceding example, the first value tells us that there is a total amount of system swap set at 2063 MB, with the second value indicating how much swap is being used (0 MB), while the third value shows the amount of swap that is still available to the system as a whole (2063 MB). So yes, based on the example data shown here, we can conclude that this is a healthy system, and no swap is being used, but while we are here, let’s use this time to discover more about the swap space on your system.
To begin, we will revisit the contents of the proc directory and reveal the total and used swap size by typing the following command:
# cat /proc/swaps
Assuming that you understand the output shown, you should then investigate the level of swappiness used by your system with the following command:
# cat /proc/sys/vm/swappiness
Having done this, you will now see a numeric value between the ranges of 0-100. The numeric value is a percentage and it implies that, if your system has a value of 30, for example, it will begin to use swap memory at 70 percent occupation of RAM.
The default for all Linux systems is usually set with a notional value between 30 to 60, but you can use either of the following commands to temporarily change and modify the swappiness of your system.
This can be achieved by replacing the value of X with a numeric value from 1-100 by typing:
# echo X > /proc/sys/vm/swappiness
Or more specifically, this can also be achieved with:
# sysctl -w vm.swappiness=X
If you change your mind at any point, then you have two options in order to ensure that no permanent changes have been made. You can either repeat one of the preceding two commands and return the original values, or issue a full system reboot.
On the other hand, if you want to make the change persist, then you should edit the /etc/sysctl.conf file and add your swappiness preferences in the following way:
vm.swappiness=X
When complete, simply save and close the file to ensure that the changes take effect.
The level of swappiness controls the tendency of the kernel to move a process out of the physical RAM on to a swap disk. This is memory management at work, but it is important to realize that swapping will not occur immediately, as the level of swappiness is actually expressed as a percentage value. For this reason, the process of swapping should be viewed more as a measurement of preference when using the cache, and as every administrator will know, there is an option for you to clear the swap by using the commands swapoff -a and swapon -a to achieve the desired result.
The golden rule is to realize that a system displaying a level of swappiness close to the maximum value (100) will prefer to begin swapping inactive pages. This is because a value of 100 is a representative of 0 percent occupation of RAM. By comparison, the closer your system is to the lowest value (0), the less likely swapping is to occur as 0 is representative of 100 percent occupation of RAM.
Generally speaking, we would all probably agree that systems with a very large pool of RAM would not benefit from aggressive swapping. However, and just to confuse things further, let’s look at it in a different way. We all know that a desktop computer will benefit from a low swappiness value, but in certain situations, you may also find that a system with a large pool of RAM (running batch jobs) may also benefit from a moderate to aggressive swap in a fashion similar to a system that attempts to do a lot but only uses small amounts of RAM. So, in reality, there are no hard and fast rules; the use of swap should be based on the needs of the system in question rather than looking for a single solution that can be applied across the board.
Taking this further, special care and consideration should be taken while making changes to the swapping values as RAM that is not used by an application is used as disk cache. In this situation, by decreasing swappiness, you are actually increasing the chance of that application not being swapped-out, and you are thereby decreasing the overall size of the disk cache. This can make disk access slower. However, if you do increase the preference to swap, then because hard disks are slower than memory modules, it can lead to a slower response time across the overall system. Swapping can be confusing, but by knowing this, we can also appreciate the hidden irony of swappiness. As Newton’s third law of motion states, for every action, there is an equal and opposite reaction, and finding the optimum swappiness value may require some additional experimentation.
Summary
In this article, we learned some basic yet vital commands that help us gauge and maintain server performance with the help of swapiness.
Resources for Article:
Further resources on this subject:
- Installing CentOS [article]
- Managing public and private groups [article]
- Installing PostgreSQL [article]