Load of Unix/Linux

Almost every engineer who has access to Unix-like operating systems knows how to view system loads. However, the working mechanism of this item may not be clear enough. I compared some related information, added my own understanding, and made some notes.

What is load? What is load average?

Load is the measure of how much the computer is working (Wikipedia: The system load is a measure of the amount of work that a computer system is doing ). In other words, the length of the process queue. Load average is the average load for a period of time (1 minute, 5 minutes, 15 minutes. [Best reference: Unix load average Part 1: How It Works]

The following is an uptime command output:

$ uptime
 18:57:48 up 423 days,  3:55,  2 users,  load average: 1.16, 1.12, 1.20

Although the definitions of various information sources are not definite. One thing you can be sure of is that you cannotPrecisionObtains the load of the current time. the minimum computing granularity is 5 seconds (calc_load is calculated once every 5Hz, 5Hz is 5 seconds, Here Hz is the system-defined variable ). refer to the Linux kernel code section:

 869        count -= ticks; 870        if (unlikely(count < 0)) { 871                active_tasks = count_active_tasks(); 872                do { 873                        CALC_LOAD(avenrun[0], EXP_1, active_tasks); 
 874                        CALC_LOAD(avenrun[1], EXP_5, active_tasks);
 875                        CALC_LOAD(avenrun[2], EXP_15, active_tasks);
 876                        count += LOAD_FREQ; 877                } while (count < 0); 878        } 879} 

 

How can I determine if the system has been over load?

For general systemsCount to judge. In the above example, if the average load is always below 1.2, and you are two "> CPU. Then there is basically no "> CPUInsufficient. That is, the average load is smaller than "> CPU.

This is a recommended Evaluation Method for Solaris Performance Tools. [I would like to recommend this book here, although it is not as detailed as I expected in load. However, the book reveals a lot of performance information. A required document for each dBA and architect .]

In this case, there are two other questions:

1 if it is multi-core "> CPU/How can I determine the hyper-threading machine?For such a machine, my advice is to see how the operating system recognizes the CPU, according to the system recognized logic "> CPUQuantity. If you want to consider the performance coefficient, we recommend that you refer to Oracle for multiple cores under different architectures "> CPU.

2. How can I determine if the application is thread-oriented?This is actually related to the M: N thread model. What is your system? Take this issue into consideration.

In most cases, if the load is too high, it may not be the same as "> CPU. There may be an exception, that isApplication scenarios. For example, use a single "> CPUMachine to create high-concurrency Web servers, the trouble is coming.

Capacity Planning)

Any relatively mature site will use tools such as cacti (rrdtool) for capacity planning. The captured load will pass the column values of 1, 5, and 15 minutes. Which of the three measurements is used? 15 minutes is the first choice [see Gunther's PPT ].

Load and system warning

Many environments with high availability requirements have established email or SMS alarm mechanisms. The setting of the load alarm threshold is not reasonable. Here we recommend the critical value (if you use tools such as Nagios to understand what this is). The upper limit is physical "> CPU(Of course, you can set a lower value than this value ). However, if it is higher than this value, it will be of little significance. For example, if the database server has 4 CPUs, an alarm should be triggered when the load is higher than 4. setting a high ratio is unlikely because someone else receives an alarm will respond to the alarm...

Misunderstanding 1: system load is always a problem with performance.

Truth: the high system load may also be due to "> CPUIntensive Computing (such as compilation)

Misunderstanding 2: system load must be "> CPUInsufficient capacity issues or quantity.

Truth: high load only indicates that too many queues need to be run. However, tasks in the queue may actually consume CPU, I/O, or other factors.

Misunderstanding 3: The system has a high load for a long time, and the CPU is preferred to be added.

Truth: load is just a representation, not a substance. Add "> CPUIn some cases, the system load is temporarily decreased, but the symptoms are not the root cause.

A little load is actually a lot of attention. In fact, the English information is comprehensive. Try to add a little new information to this article. Please confirm or inform me of any unreasonable reason for writing or any objection.

-- "> EOF--

 

"> FAQ1: Database Server suddenly "> CPU100% busy. What should I do?

 

A: Generally, this is caused by poor "> SQLCause. We recommend that you capture the slow query log. For "> SQL. Based on experience, each "> CPUCore can process 100-400 mb of data in one second. If it is a large number of concurrent I/O operations, although the storage throughput may not be so large, it may also take "> CPU"Full ".