Linux performance optimization 13th: performance analysis concepts

Talking about linux performance tuning 13: performance analysis concepts; talking about linux performance tuning 12: Red Hat optimization strategy Tuned http://www.2cto.com/os/201307/229281.html About the performance metrics of the processor .? CPU usage [CPUUtilization] this may be the most direct indicator... talking about linux performance tuning 13: performance analysis concepts talking about linux performance tuning 12: Red Hat optimization strategy Tuned http://www.2cto.com/os/201307/229281.html About the performance metrics of the processor.? CPU usage [CPU Utilization], which may be the most direct indicator, indicates the overall usage of each processor. In the IBM System x architecture, if the CPU usage exceeds 80% for a period of time, it may indicate a CPU bottleneck.? User Time indicates the CPU usage of the User process, including the Nice Time. When the user's time value is very high, it indicates that the system is performing the actual work.? System Time indicates the CPU usage of kernel operations, including hard interrupt [IRQ] and soft interrupt [SoftIRQ]. A high system time value indicates that a network or drive Stack may have a bottleneck. Generally, the system only spends a little time on the kernel.? Waiting for [Waiting] the total CPU time required to wait for I/O operations. similar to blocking [Blocked], the system should not spend too much time Waiting for I/O operations; otherwise, you should check the performance of the I/O subsystem.? Idle time indicates the percentage of Idle CPU.? Nice time indicates the CPU usage of the re-nicing process.? Average Load average: the average Load is not a percentage. it is the average of the sum of the following values:-number of processes waiting for execution in the queue-number of processes waiting for the completion of an uninterruptible task. That is, the average value of the sum of TASK_RUNNING and TASK_UNINTERRUPTIBLE. If the process requesting CPU time is blocked (), the average load will rise. On the contrary, if each process can be executed immediately, the CPU cycle will not be missed, and the average load will decrease.? Runable processes indicates the process to be executed. This value should not exceed 10 times the number of physical processors for a period of time, otherwise the CPU may have a bottleneck.? Blocking [Blocked] The process cannot continue until the I/O operation is complete. Process congestion may cause a bottleneck in I/O.? Context switch: The number of processes in the system that are exchanged. Too many context exchanges are related to a large number of interruptions, which may imply a problem with the drive or application. Generally, context switching is not required because you only need to refresh the CPU cache each time, but some context switches are necessary.? Interrupts: the number of Interrupts, including hard and soft Interrupts. Hard interruptions have a very negative impact on system performance. The high interrupt value indicates that the software has a bottleneck, which may be the kernel or driver. Remember that the interrupt value also includes the interruption caused by the CPU. Memory performance indicators.? Idle memory [Free memory] does not have to worry too much about idle memory value compared with other operating systems. As described in 1.2.2 "virtual memory management", The Linux kernel allocates a large amount of unused memory for file system cache usage. Therefore, the amount of memory used for caching and caching is deducted to obtain the actual idle memory.? The value of Swap usage indicates the number of Swap spaces in use. As described in 1.2.2 "virtual memory management", the use of swap space can only tell you how effective Linux is in memory management. To determine whether the memory is stored In the bottleneck, the number of Swap In/Out is used. If Swap In/Out remains above 200 to 300 pages per second for a long time, it may indicate a memory bottleneck.? Buffer and cache is used as the cache for file systems and block devices? Slabs indicates the memory used by the kernel. Note that the kernel page cannot be switched to the hard disk.? Active and inactive memory (Active versus inactive memory) provides information about Active memory. The inactive memory will be switched to the hard disk by kswapd as a candidate. For more information, see "page frame collection" network performance indicators.? Packets received and transferred [Packets received ed and sent] can this indicator tell you the number of Packets received and sent by a specific network adapter? Bytes received and sent Bytes indicate the number of Bytes received and sent by the specified network adapter.? Number of conflicts per second (Collisions per second) this value provides the number of network conflicts that occur on the specified Nic. Conflicting values indicate bottlenecks in the network architecture rather than servers. In most correctly configured networks, conflicts are rare unless the network architecture is composed of hubs.? The number of dropped Packets (Packets dropped) by the kernel may be due to firewall configuration problems or lack of network buffer? Overruns indicates the number of times the network interface is buffered. This indicator can be used with the number of discarded packets to determine whether the bottleneck is from the network buffer or the length of the network queue.? Error [Errors] indicates the number of failed frames. This is often caused by network mismatch or some network cable damage. Damage to some Gigabit network cables of copper cables may cause serious performance problems. Performance indicators of block devices.? IO wait [Iowait] the time the CPU spends waiting for I/O operations. If this value continues very high, it is likely that I/O has a bottleneck.? Average queue length [Average queue length] number of I/O requests. Generally, the hard disk queue value ranges from 2 to 3, which may indicate a bottleneck in hard disk I/O.? Average wait time [Average wait] the Average time spent by the I/O request service. The waiting time includes the actual I/O operation time and the waiting time in the I/O queue. The unit is ms.? Number of Transfers per second (Transfers per second) indicates the number of I/O operations performed per second (including reading and writing ). Combined with the number of bytes transmitted per second (kBytes per second), it can help determine the average system transmission size. The average transmission size is usually the same as that of the hard disk subsystem.? Number of read/write Blocks per second [Blocks read/write per second] this indicator indicates the number of read/write Blocks per second. in the 2.6 kernel, the block size is 1024 bytes, earlier kernels may have different block sizes, from 512 bytes to 4 kB.? The number of read and write bytes per second (Kilobytes per second read/write) indicates the actual number of data read and write by block devices. the unit is KB.