Detailed parsing of AIX performance monitoring Topas commands

After executing the TOPAS command: #topas

Zone 1: Reflect CPU usage and working conditions

Kernel: CPU time ratio used by the kernel of the operating system.

Operating system as the basic software, for the application support and service, the operation of itself also requires a certain amount of CPU and memory resources (by the way the memory resources, no longer elaborated on this content), especially memory resources, the heavier the system load, the corresponding kernel consumes more CPU and memory resources. In general, the kernel consumes too much CPU time. Generally less than the CPU usage of the app.

User: The CPU time ratio consumed by the process.

This is the key value for CPU usage. This usage reflects the sum of CPU time ratios used by the various software users running on the basis of the operating system. In general, if the User+kernel is more than 70% consecutive, you can assume that the system may have a problem with the severity of the CPU.

The WAIT:CPU is in the waiting state for the CPU time ratio.

CPU wait is generally waiting for IO response, it is known that the main bottleneck of the current computer is in IO. When the application executes, it needs to read and write data from external storage such as disk, and the process will initiate an IO request and wait for IO to complete. This waiting process consumes CPU time and is wait. When this value is high, it means that Io is too late to respond to a lot of IO requests, and at this point it can only be optimized from the IO level.

Idle:cpu idle time ratio, this is needless to say. Is how much CPU time ratio is idle.

The main possible cause of a problem with CPU usage is that the database server executes a SQL or stored procedure (the stored procedure is a packaged SQL package) and requires a large number of operations (typically unreasonable for software design). Or where there is an exception in the application, such as a dead loop, or another logic error in writing the program. A general program error causes a CPU to be fully occupied, such as the above 20% occupancy because a trading program takes up a full time slice of CPU (the system has a total of 5 CPUs).

Zone 2: Status of Network utilization

Netwok; Lists the NIC interface, Kbps is the number of kilobytes per second (Kbytes) I-pack per second, o-pack the number of packets output per second kb-in bytes entered per second kb-out bytes output per second.

　　when we find the network congestion (the error occurs when the NIC transmission fails, that is, the NIC sends the packet failed.) Or if the network response is significantly slower, if the CPU is not a problem, then check the network traffic) to find that a network card of Kbps continues to be greater than four digits, or even five digits (this value if the NIC gigabit or hundred trillion). It is necessary to look at this network card is what network card, in dealing with what business . In the command line execution netstat–in view the corresponding en* interface IP address, through the IP address to see whether it is with the official network card or production service network card high traffic. Then through the Netstat–v en* look at the details of the network card work status, how many error packets, conflict packets, CRC checksum error or network reset over the information. For the above information, please see the output of netstat–v en* in detail. If there is a large number of CRC, the wrong packet, the network cable may have problems or poor contact.

If these are normal, and the network response is slow, it is possible that the switch is congested.

Possible causes of problems in the network: loading large amounts of data via a Gigabit pipe network (previously), large queues of long-time FTP transmissions, or network cables, switch issues, etc.

Zone 3: Status of disk usage

Disk busy% The percentage of disks that are busy, that is, the maximum IOPS (IO operations per second) that the disk can meet and the current number of IO. Other parameters are no longer explained. Words too literally can be.

Generally see the disk busy%, when the disk busy% is more than 85%, that is, the disk is considered busy, it is possible to have problems. Of course, you know that the content that has been determined to generate a lot of IO operations doesn't matter, so it can be done.

The cause of the problem: the application Server write log process or query log process a large number of read and write logs, resulting in high disk busy rate, or other programs frequently read and write disk caused. In the system, the HDISK0,HDISK1 is generally a system disk, and the relative iops of the built-in SCSI disk is low. It is easy to run at full load.

Zone 4: Status of information reflecting processes

Name: The names of the processes, that is, the names of the binaries that are started when the process is executed.

PID: The ID of the process, the ID of the process is unique in the system, is we understand the tracking process information important values. Trace CPU usage of the process, disk IO read/write, process memory and pagingspace, etc. all need to be used.

CPU%: The rate at which the process consumes CPU time.

PGSP: The amount of space pagingspace the process occupies.

Owne: The owner of the process, that is, which operation user initiated the process.

In Topas, the default is to list the top CPU-intensive process information for reference, if the previous first area of the CPU usage continues to be high, it is necessary to look at this is the process consumes a lot of CPU resources, to see which user's process, if they do, then kill or find the project team to solve.

Zone 5: Status of Memory pages and paging space information

The paging space, the space on the disk, is used for memory space in the AIX operating system. The specific theory is no longer elaborated, please refer to the operating system content for more information. The speed of disk space is of course more than 10 times times slower than memory. So, just a memory page of a temporary storage, storage or those long-term not how to use the memory page. If there is a lot of paging, this time there is trouble, stating: Memory is not enough!

The area is mainly concerned about pagein,pageout if both values are greater than three digits, and long-term greater than this value, is technically called the memory bump, that is, the memory page is constantly swapped to disk space, and from disk space to read the memory page, the system's memory usage efficiency is very poor, System response performance is also slowing down.

This information can also be used in Vmstat, the PI and PO columns correspond to here. Of course, if there is only page out, or only page in, or a short time some of the page swap out, there is no problem, focus on it.

Zone 6: Information that reflects memory usage

REAL,MB: The total amount of memory that the operating system actually owns, in megabytes.

%COMP: Compute-based memory occupancy ratio

%noncomp the ratio of non-computed memory occupancy.

%client is also a non-computational type of memory, Noncomp includes client-type memory, the JFS file system uses memory of Noncomp, in order to differentiate, JFS2 and NFS use memory as client.

Computational memory is the actual memory used by the process, for example, when we write the program malloc memory, or use the stack in the sort, the variables in the process will need to be stored in memory, this memory is calculated memory (not exhaustive, for reference only). While the operating system in the file read and write, the required IO buffer, or when we write the program, open files, read and write files, both in the file buffer. (Bare device exception, CCCC database uses RAC, data storage all use bare device, on the database server, the data file buffer in the Oracle's SGA area in data buffer (this area system is considered to be computational memory), does not occupy non-computational memory. )

There are many possible causes of memory problems. The main: The process uses more memory, for example, the CCCC database Server A large number of Oracle connections use a lot of memory, or a database executed in a SQL script or the execution of a stored procedure requires a large amount of memory to complete its operation (the case in the exception library has occurred, The execution of a stored procedure caused the operating system memory is exhausted, the PG is also exhausted, the operating system automatically executes Pgsp_kill, the process to kill, I am also the first to know the AIX system and this function, hehe). The second major problem is the memory leak, the simplest memory leak, is the application of memory space, after use is no longer used, but also not released. When we wrote the program, malloc was not free. This leads to serious problems, with the execution of the program, the availability of less physical memory, the end of the hang, and the regular restart of the application to solve.

The memory-paging mechanism of the operating system causes the memory pages that are not used in the program to end up on the PG, and the paging space will continue to grow. This is how the system problem is caused by the application.

Zone 7 reflects the usage of the paging space

If the usage of the paging space increases for a long time, it indicates that the system is out of memory, has started to use disk space to buffer memory, if the PG usage continues to grow, or greater than 50%, need to be vigilant (to 50% in the monitoring platform is already the main alarm!), and immediately submit the system administrator to analyze the reason for memory growth. If the value continues to grow, the system will definitely hang up!

---organized from the network

Detailed parsing of AIX performance monitoring Topas commands