Assessing Groundwork Monitoring Server Performance

Version 3 by NotSupportContact-Mark Carey
on Aug 04, 2011 15:31.
<< Changes from 2 to 3
compared with
Current by NotSupportContact-Mark Carey
on Aug 04, 2011 15:32.
Key
This line was removed.
This word was removed. This word was added.
This line was added.

Changes (20)

View Page History
...

h2. Introduction

This document is intended to help you assess your GroundWork Monitoring Server's performance. As the server's workload increases towards its upper range, certain functions will begin to manifest problematic behavior.

h2. Typical First-Noticed Symptoms

# One typical symptom of a over-loaded server is gaps in RRD Performance Graphs. As check latencies increase, data points may start falling outside the allowed tolerances and be dropped from the graphs.
# Changes made in the Monarch 'Configuration' Tab take noticeably longer to propagate from a Monarch 'commit' to the GroundWork Status Viewer pages.

h2. Nagios Engine Program-Wide Performance Information (Check Latency)

The first place to look for indications that your system is over-loaded is the Nagios Performance CGI. From the GroundWork Portal, click on 'Monitoring Server' option. This will bring you to a Nagios display of the Monitoring Server and its services. In the secondary menu, you then pick 'Performance Information'.

...

h2. Nagiosstats

[Nagiosstats|http://nagios.sourceforge.net/docs/2_0/nagiostats.html] is a command-line version of the CGI shown above. It is simpler and has less dependencies. It also carries slightly different information, such as 'Total Services' and 'Total Hosts'
{code}
...

h3. The Linux Utility, [top|http://unixhelp.ed.ac.uk/CGI/man-cgi?top]

Another assessment of performance is CPU Load. This metric is expressed in how many processes are in a runnable state averaged over 1 minute, 5 minute and 10 minute intervals. If this number is, say 4, that means that there were 4 runnable processes on average over the minute intervals. The following [top|http://unixhelp.ed.ac.uk/CGI/man-cgi?top] output shows load averages of 0.58, 0.63, 0.64. These numbers state that about half of the time there were no processes in a runnable state.
[Top|http://unixhelp.ed.ac.uk/CGI/man-cgi?top]'s main strength is in showing you which processes are in what states. Which processes are taking the most resources? Top can show you. A typical scenario is where the Java listener that accepts feeds of data and posts them to the Collage DB, and the MySQL daemon, are the most active processes.
...

h3. The Linux Utility, [uptime|http://unixhelp.ed.ac.uk/CGI/man-cgi?uptime]

[Uptime|http://unixhelp.ed.ac.uk/CGI/man-cgi?uptime] is a smaller footprint tool that you can use to get the same load averages that [top|http://unixhelp.ed.ac.uk/CGI/man-cgi?top] had.
{code}# uptime
...

h3. The Linux Utility, [vmstat|http://unixhelp.ed.ac.uk/CGI/man-cgi?vmstat]

[Vmstat|http://unixhelp.ed.ac.uk/CGI/man-cgi?vmstat] has many options to it. Each option may output totally different types of metrics. One of the most useful is [vmstat|http://unixhelp.ed.ac.uk/CGI/man-cgi?vmstat] \-s.
{code}[root@lunias ~]# vmstat -s
...

h3. Linux utility: [sar|http://perso.orange.fr/sebastien.godard/man_sar.html]

[Sar|http://perso.orange.fr/sebastien.godard/man_sar.html] is the system activity reporter. By interpreting the reports that [sar|http://perso.orange.fr/sebastien.godard/man_sar.html] produces, you can locate system bottlenecks and suggest some possible solutions to those annoying performance problems. The Linux kernel maintains internal counters that keep track of requests, completion times, I/O block counts, etc. From this and other information, [sar|http://perso.orange.fr/sebastien.godard/man_sar.html] calculates rates and ratios that give insight into where the bottlenecks are.
The key to understanding [sar|http://perso.orange.fr/sebastien.godard/man_sar.html] is that it reports on system activity over a period of time. You must take care to collect [sar|http://perso.orange.fr/sebastien.godard/man_sar.html] data at an appropriate time (not at lunch time or on weekends, for example).
...

h4. Other [sar|http://perso.orange.fr/sebastien.godard/man_sar.html] Utilities

* [iostat|http://perso.orange.fr/sebastien.godard/man_iostat.html]
* [sadc|http://perso.orange.fr/sebastien.godard/man_sadc.html]
...

h4. Installing & Configuring [sar|http://perso.orange.fr/sebastien.godard/man_sar.html] (Sysstat)

* [http://www.linuxfromscratch.org/blfs/view/svn/general/sysstat.html]

h2. RRD Performance Data Collection

GroundWork can collect performance data using an eventhandler. It runs after each check finishes, reading a DB Table and trying to log performance data. Performance data collection can contribute to a slow system.
The simplest way to gauge this is to count the RRD files in /usr/local/groundwork/rrd that have been modified that day.
...

h2. GroundWork Subsystems

There are various log files and other audit aids in GroundWork Monitor that can give you hints about how well the server is performing.

h3. Process Performance Data Event Handler Log

For the purposes of this discussion, we will want to focus on the eventhandler logfile, /usr/local/groundwork/nagios/var/log/process_service_perfdata_file.log.

...
{code}
First, this debug_level variable only needs to be set to a value above 1 when you are actively debugging something in the Performance data collection subsystem. Otherwise, it should be set to 0 or 1.

The contents of the log can be illuminating. This is a special-purpose eventhandler, one that runs after each service check that has performance data handling enabled - that's the default. Many checks, though, don't have a metric - such as any check that returns true or false. To tune the system, you should disable performance data collection for those services whose results aren't suited for such.

Also, just because you could plot the data on a graph doesn't mean that you should. Conservation of effort here will pay off. If no one is asking for the data, and you aren't particularly interested either, then turn it off.

To turn off performance data handling on a particular host or service, go into the Service record in the Monarch 'Configuration' Tab and, in the 'Service Detail' Tab, uncheck the 'Process Perf Data' property. In the 'Host Detail' Tab it is called the 'Process performance data' property.

Executing the Process-Perf-Data eventhandler only for services that actually are set up for performance data collection and not others, and turning off the logging function when not actively tuning the system, will save computing resources.

h2. GroundWork Diagnostics Tool

GroundWork Customer Support has created a tool that methodically collects the above information and more. You could run the GroundWork Diagnostics Tool and send the information to Support or use it in your own analysis. See the kb article for more information: [Running the gwdiags diagnostic tool]



h2. Conclusions

Having assessed your system and tuned it as well as possible, you might find that you still have a performance issue. Contact [Support|https://cases.groundworkopensource.com], or your GroundWork Account Representative to explore other options.