본문 바로가기

Linux

Linux Performance Metrics


출처 : http://support.uptimesoftware.com/article.php?id=117#6

The up.time Linux agent collects the following performance metrics from the systems on which it is installed:

Each set of performance metrics is averaged over an interval of one second.

CPU

The up.time agent uses the sar -urWqR 1 command to compare the system counters during a one-second interval. The statistics returned by the agent are averaged for all CPUs on the system.

MetricExplanationSource
% USRThe percentage of time that the processor spends in user mode (a processing mode for applications and subsystems)./proc/cpuinfo
% SYSThe percentage of time that the kernel spends processing system calls./proc/cpuinfo
% WIOThe amount of waiting time that a runnable process for a device takes to perform an I/O operation./proc/cpuinfo
% TotalThe total amount of User %, System %, and Wait I/O %/proc/cpuinfo
Run Queue LengthThe percentage of time that one or more services or processes are waiting to be served by the CPU./proc/cpuinfo

Multi-CPU

The up.time agent uses the sar and mpstat utilities on a Linux system to collect the metrics in the table below from Linux systems with multiple CPUs. The agent averages the statistics from each CPU using the sar -x SELF -I SUM -P ALL -wu 1 command, which compares the system counters during a one-second interval. The statistics that the agent returns are for the entire system, per CPU.

MetricExplanation
User %The percentage of CPU user processes that are in use.
System %The percentage of CPU kernel processes that are in use.
Wait I/O %The percentage of time that a process which can be run must wait for a device to perform an I/O operation.
SMTXThe number of read or write locks that a thread was not able to acquire on the first attempt, as reported by the mpstat command.
XCALThe number of interprocess cross-calls. In a multi-processor environment, one processor sends cross-calls to another processor to get that processor to do work. Cross-calls can also be used to ensure consistency in virtual memory. Heavy file system activity (such as NFS) can result in a high number of cross-calls.
InterruptsThe number of CPU interrupts.
Total %The total amount of User %, System %, and Wait I/O%.

Memory

The up.time agent uses the sar -urWqR 1 command to collect memory metrics from a Linux system. The command also compares the system counters during a one-second interval. The statistics that the agent returns are for the entire system.

MetricExplanationSource
Free MemoryThe amount of physical memory available to the operating system, system library files, and applications./proc/meminfo
Cache Hit RateHow often the system accesses the CPU cache./proc/meminfo
PageOut per SecondThe rate at which pages were written to disk./proc/meminfo
PageIn per SecondThe rate at which pages were read from or written to the disk./proc/meminfo
PageFree per SecondThe number of pages that are freed from memory each second./proc/meminfo
PageScan per SecondThe average number of pages that are scanned each second./proc/meminfo
Free SwapThe amount of available free swap space, as a percentage of total available free swap space./proc/meminfo

Disk

The up.time agent gathers file system statistics for each file system using the df -lk command. Disk statistics (e.g. %busy, reads per second and writes per second) are output per disk and compared between polling intervals using the iostat -d -x 1 2command.

MetricExplanationSource
Disk (Spindle) NameThe names of each disk on the system./proc/diskstats
Usage (% Busy)The percentage of time during which the disk drive is handling read or write requests./proc/diskstats
Throughput (Blk/s)The number of read and write operations on the disk that occur each second./proc/diskstats
Read/Writes/sThe average number of bytes that have been transferred to or from the disk during write or read operations./proc/diskstats
Average Queue LengthThe number of threads that are waiting for processor time./proc/diskstats
Average Service TimeThe average amount of time, in milliseconds, that is required for a request to be carried out./proc/diskstats
Average Wait TimeThe average time, in milliseconds, that a transaction is waiting in a queue. The wait time is directly proportional to the length of the queue./proc/diskstats

Network

The up.time agent uses the netstat -s command to retrieve a combined total of TCP Retransmits for all network interfaces. Other network statistics (e.g. kbps, errors and collisions) are averaged, per interface, using the sar -n DEV -n EDEV 1 command, which compares the system counters during a one-second interval.

MetricExplanationSource
In KbpsThe rate, in kilobytes per seconds, at which data is received over a specific network adapter./proc/net
Out KbpsThe rate, in kilobytes per seconds, at which data is sent over a specific network adapter./proc/net
In ErrorsThe number of inbound packets that contained errors, which preventing those packets from being delivered to a higher-layer protocol./proc/net
Out ErrorsThe number of outbound packets that could not be transmitted because of errors./proc/net
CollisionsThe number of signals from two separate nodes on the network that have collided./proc/net
TCP RetransmitsThe number of packets that have been re-sent over a network interface. The agent returns a combined total for all interfaces./proc/net

Process

The up.time agent uses the ps -eo command to collect the process information listed in the table below from a Linux system. By default, the agent gathers the top 20 processes and sorts them by the highest CPU usage.

MetricExplanationSource
PIDThe unique identifier of a specific process./proc/stat
PPIDThe identifier of the process that the process that is currently running./proc/stat
UIDA value that identifies the current user./proc/stat
GIDA value that identifies a group of users./proc/stat
Memory ConsumedThe amount of memory that is being used by a process./proc/stat
RSSThe amount of physical memory that is being used by a process./proc/stat
CPU % Utilization by ProcessThe percentage of CPU time that is being used by individual processes./proc/stat
Memory % Utilization by ProcessThe amount of physical memory that is being used by individual processes./proc/stat
Process Start TimeThe time at which the process started./proc/stat
Process Run TimeThe time at which the process started./proc/stat
Number of Processes RunningThe total number of processes that are currently running on the system./proc/stat
Number of Blocked ProcessesThe total number of processes that are blocking resources./proc/stat
Number of Waiting ProcessesThe total number of processes that are waiting to be executed by the CPU./proc/stat
Execs per SecondThe total number of system calls that are executed each second./proc/stat
Process Creation RateThe total number of processes that are being spawned over a specified time period./proc/stat

Workload

The up.time agent uses the ps utility to collect workload information from a Linux system. Workload statistics (based on the same 20 processes that were gathered from the Process method) are sorted within up.time's core. The workload processes that the agent gathers include the user/group/process name and their invividual statistics, which can be sorted based on the user's desired graph presentation (e.g. user, group or process name).

MetricExplanationSource
Workload by ProcessThe demand that network and local services are putting on a system, based on the processes that are running./proc/load
Workload by UserThe demand that network and local services are putting on the system, based on the IDs of the users who are logged into a system./proc/load
Workload by GroupThe demand that network and local services are putting on the system, based on the IDs of the user groups that are logged into a system./proc/load
Workload Top 10 by ProcessThe 10 processes that are consuming the most CPU resources./proc/load
Workload Top 10 by UserThe 10 processes the are consuming the most CPU resources, based on user ID./proc/load
Workload Top 10 by GroupThe 10 processes the are consuming the most CPU resources, based on group ID./proc/load

User

The up.time agent uses the following commands to collect user statistics from a system:

  • ps -eo
  • last | head 10 (login history for the last 10 users on the system)
  • who (lists who is currently logged into the system)
MetricExplanation
Login HistoryThe number of times or frequency at which a user has logged into a system during any 30 minute time interval.
SessionsThe number of sessions or number of distinct users who are logged into a system during any 30 minute time interval.

'Linux' 카테고리의 다른 글

Byobu 도움말  (0) 2012.07.24
[Linux/Unix]디스크 쿼타(Disk Quota)  (0) 2012.07.20
시스템 성능향상을 위한 팁!!  (0) 2012.07.04
SAR in Ubuntu  (0) 2012.06.22
crontab  (0) 2012.06.21