订阅内容

Performance analysis of computer systems is often considered black art. It requires significant experience with the operating system, applications, networking and diagnostic tools. This is a vast amount of ground to cover and it is nearly impossible for anyone to know it all (though I am fortunate to have met many Red Hatters who seem to have mastered all of the above!).

As if it were not difficult enough, most operating systems have their own diagnostic tools and utilities to extract performance information. Yes, there are many common tools that have been ported to various platforms, but they are often quirky with inconsistent output or parameters. I’m referring to vmstat, iostat, netstat, tcpdump, etc. The SysStat (SAR) tools and service brought us close to a unified and standardized service that collects almost all of the data on a regular basis. However, it has limits regarding the granularity of collected data and can not capture per-process metrics at all. Most frustrating is that the data is collected in binary format that becomes incompatible on other platforms and versions. While the data can be converted, it can be a significant hurdle under pressure at 3:00 a.m. or to novice users.

Enter Performance Co-Pilot (PCP), a system performance and analysis framework that is modularized and easy to extend. Numerous plugins and APIs are available to integrate the data collection into your custom monitoring solution or other common monitoring tools. There is multiplatform support for VMware, Windows and Mac, as well as agents for common applications and devices such as Oracle DB, PostgreSQL, Cisco devices and much more.

Beginning with versions 6.5 and 7.0 of Red Hat Enterprise Linux, PCP is included in the base software channels. Going forward it is intended as standard tool to provide performance related diagnostic metrics.

 

Overview

Without going too deep into the theory, PCP has three primary parts:

Performance Metric Domain Agents (pmda)

These are the various agents, collecting data from the given domain or subsystem (kernel, memory, storage, DB application, etc).

Performance Metric Collection Daemon

This is the service daemon that controls the various pmdas and acts as a router for clients to request data from the service.

Clients

These are the various clients that would request the data in order to log to storage (local or remote) or present to a user via a graphical or command line interface.

 

Getting started

To get started collecting metrics as a service (similar to the SAR/Sysstat service), just type the following simple commands in Red Hat Enterprise Linux 7:

 

yum install pcp

systemctl enable pmcd

systemctl start pmcd

systemctl enable pmlogger

systemctl start pmlogger

 

Next, you can test various commands that are very similar to other utilities you are probably familiar with in order to see an overview of system activity.

 

pmstat like vmstat

pmiostat like iostat

pmatop like top enhanced

 

Here is a basic example of pmatop

pmatop2.png

 

Or go deep examining thousands of precision metrics

pmval -t 1 -s 3 network.interface.in.bytes


metric:    network.interface.in.bytes

host:      rhel7

semantics: cumulative counter (converting to rate)

units:     byte (converting to byte / sec)

samples:   3

interval:  1.00 sec


             eth0                    lo               docker0    

            118.0                   0.0                   0.0    

             65.97                  0.0                   0.0    

            118.0                   0.0                   0.0    

 

Because it runs as a service like SAR/SysStat (with up to 1 second precision* unlike SAR/Sysstat), you can examine historical data to help you troubleshoot prior performance.

 

pmval -a ./20161129  -t 30 -S "Nov 29 08:00" -T "Nov 29 08:05"  network.interface.in.bytes

metric:    network.interface.in.bytes

archive:   ./20161129

host:      rhel7

start:     Tue Nov 29 08:00:00 2016

end:       Tue Nov 29 08:05:00 2016

semantics: cumulative counter (converting to rate)

units:     byte (converting to byte / sec)

samples:   11

interval:  30.00 sec

                         eth0                    lo               docker0    

08:00:30.000              26.53                  0.0                   0.0    

08:01:00.000              26.00                  0.0                   0.0    

08:01:30.000              26.00                  0.0                   0.0    

08:02:00.000              26.00                  0.0                   0.0    

08:02:30.000              26.00                  0.0                   0.0    

08:03:00.000              26.00                  0.0                   0.0    

08:03:30.000              27.33                  0.0                   0.0    

08:04:00.000              28.20                  0.0                   0.0    

08:04:30.000              26.87                  0.0                   0.0    

08:05:00.000              26.00                  0.0                   0.0   

 

 

Graphical representation

PCP also excels over the traditional SysStat/SAR tooling with multiple graphical presentation options. The client app pmcharts (or PCP Charts) act as a local client program that can present data collected from the local host, a remote host, or archive files. Numerous default reports are provided and the Overview report gives a nice dashboard of the system. While the graphics are not the most elegant, they are very functional and and easy to use for analysis.

 

pcp-charts.png

 

There are a few more elegant graphical options including “Vector” (https://github.com/Netflix/vector) which was contributed by our friends at Netflix. Installing the following packages will give you multiple frameworks to experiment with.

yum install pcp-webjs pcp-webapp-vector pcp-webapp-grafana pcp-webapp-graphite

 

After installing, point your browser to the landing page (http://localhost:44323/index.html) which provides links to each of the frameworks and a nice summary. You can click on the link for Vector to see its excellent default dashboard.

pcp-vector.png

 

As you dig into the many options and metrics available, you will quickly see that there is a tremendous amount of performance information at your fingertips. And the PCP website provides multiple case study examples of analysis to help you get started using the tool.

 

profile3.jpgTerry Bowling is a Red Hat TAM in the NA Central region and is transitioning to Technical Product Manager. He has been migrating workloads from UNIX to Linux since 1998 and has supported business environments for major telecom and pharmaceutical companies. Most recently he has been focused on enabling our customers to migrate to container platforms using Red Hat Atomic/OpenShift, as well as adopting SAP HANA.

Twitter: https://twitter.com/terry_bowling #RHTAM

A Red Hat Technical Account Manager (TAM) is a specialized product expert who works collaboratively with IT organizations to strategically plan for successful deployments and help realize optimal performance and growth. The TAM is part of Red Hat’s world class Customer Experience and Engagement organization and provides proactive advice and guidance to help you identify and address potential problems before they occur. Should a problem arise, your TAM will own the issue and engage the best resources to resolve it as quickly as possible with minimal disruption to your business.


关于作者

Terry Bowling has been designing and working with customers on UNIX and GNU/Linux environments since 1999. He brings this experience to the RHEL Product Management team to provide the best experience to assembling and deploying RHEL for customers. This includes the RHEL installer, image builder and related build services for RHEL being developed at Console.RedHat.com.

Read full bio
UI_Icon-Red_Hat-Close-A-Black-RGB

按频道浏览

automation icon

自动化

有关技术、团队和环境 IT 自动化的最新信息

AI icon

人工智能

平台更新使客户可以在任何地方运行人工智能工作负载

open hybrid cloud icon

开放混合云

了解我们如何利用混合云构建更灵活的未来

security icon

安全防护

有关我们如何跨环境和技术减少风险的最新信息

edge icon

边缘计算

简化边缘运维的平台更新

Infrastructure icon

基础架构

全球领先企业 Linux 平台的最新动态

application development icon

应用领域

我们针对最严峻的应用挑战的解决方案

Original series icon

原创节目

关于企业技术领域的创客和领导者们有趣的故事