How can I speed up PRTG—especially for large installations?

Votes:

2

Your Vote:

Up

Down

For a few hundred sensors PRTG usually doesn't show any speed and performance problems. But with thousands of sensors one may experience excessive CPU load, slow web interface or sensors that are not scanned at the specified intervals.

So what can I do to make PRTG work faster, use fewer system resources, work more reliable?

  • What things are to consider before installation?
  • What are "best practices" for large installations of PRTG?
  • What can I do to make the user experience faster, especially the web interface?
  • What can I do to speed up my installation?

hardware important installation make-faster planning prtg sensor-type server-settings speed troubleshooting

Created on Mar 19, 2010 12:24:20 PM by  Daniel Zobel [Paessler Support] (21,383) 3 3

Last change on Mar 31, 2011 10:38:15 AM by  Daniel Zobel [Paessler Support] (21,383) 3 3



4 Replies

Accepted Answer

Votes:

2

Your Vote:

Up

Down

This article applies to PRTG Network Monitor 8 and 9
Most content is also true for PRTG Network Monitor 7

How to Speed Up PRTG

There a several areas you can consider for speeding up your PRTG system, regarding the following topics:

  1. Hardware Considerations
  2. Operating System Considerations
  3. PRTG Server Settings
  4. Sensor Type and Monitoring Considerations
  5. Housekeeping and Maintenance
  6. More Resources and Facts


Note: In order to test the speed of your web interface, log in to the PRTG web interface and manually call the /speedtest.htm URL in your browser, so you can measure the success of any optimization efforts. For example:

http(s)://**YOUR_PRTG_IP_OR_DNS**/speedtest.htm


1. Hardware Considerations


Choose performant hardware.

Real Servers are Better than Virtual Servers

Especially when running installations with a great number of sensors (beyond 1,000), real hardware often performs better than virtual environments. We have seen very performant VMs that could not beat real hardware (due to network and disk speed bottle necks). So if your system slows down on a virtual server, try moving PRTG to real hardware.

We strongly recommend to use "bare metal" for installations of 5,000 sensors or more, and regard it mandatory for 10,000 sensors and more. In these cases you should run at least the core server on real hardware, remote probes can be installed on virtual machines.

Use Multi-Core, Multi-Threaded CPUs and/or Multi Processor Systems, prefer Intel

PRTG makes heavy use of multi threading, so prefer systems with multiple cores. In our tests, Intel processors outperformed AMD CPUs.

Install more RAM memory

More available memory can significantly improve the speed of PRTG.

  • 32-Bit OS: Install 4 GB of RAM memory. This will maximize the available RAM and disk cache.
  • 64-bit OS: Install at least 4 GB of RAM memory, 8 GB is better. PRTG itself can use up to 4 GB RAM. The rest is for OS itself and its disk caching

Use a Fast Disk Sub System

Remember that no RAID is faster than RAID 1 / 5 / 10.

As a bonus, you can use SSD solid state disk drives. We have not yet extensively tested this, but early measurements show that using an SSD disk in a PRTG server can lead to 25-35% faster page load times compared to a normal disk drive.

More

For detailed test figures on different hardware platforms and operating systems, please see "More Resources" section below.


2. Operating System Considerations


There are big differences between the Windows versions.

64-Bit Windows Version is Better Than 32-Bit

According to our tests, 64-bit operating systems have a somewhat higher performance compared to 32-Bit systems.

Select the Fastest Operating System Available to You

Best operating systems for PRTG, in order of performance:

  1. Windows Server 2003 R2
  2. Windows Server 2008 R2
  3. Windows XP
  4. Windows 7

Avoid Windows Vista and Windows 2008 R1 At All Cost

Both Vista and the first edition of Windows Server 2008 have serious performance issues, especially when using WMI. Do not use them for large installations!

Avoid the "Balanced" Power Plan

On Windows 2008 the power plan "Balanced" is the default setting. Switching to the "High Performance" power plan can lower your average web page times by 30-50%, also on other Windows versions! Our observation was that the effect is even stronger when the processor has many cores.

More

For detailed test figures on different hardware platforms and operating systems, please see "More Resources" section below.


3. PRTG Server Settings


You can win a lot by a smart configuration of your PRTG server!

Use the Latest PRTG Version

We continually improve our software. To profit from all speed enhancements, you should always run the latest PRTG version available for download.

Configure Memory Usage

Check the memory usage of your PRTG Server process. For values beyond 1 GB use the PRTG Server Administrator program (Memory usage tab) to reduce RAM memory usage. Choose the smallest settings that still make sense for your use case. The objective is to avoid slow memory swapping at all cost.

Memory Usage Settings in PRTG Server Administrator

Switch on "Speed" Option

Available for PRTG Network Monitor version 7.3.3 or later

This setting disables a few non-mission-critical features to speed up the web interface: From the PRTG 8 main menu, choose Setup | System Administration | System & Website (for PRTG 7, this is Setup | System Setup). In the Performance Strategy settings, choose the More Speed option.

For detailed information, please see What does PRTG's "More Speed" option do?

Switch PRTG's Web Server from HTTPS to HTTP

If you can afford a lower security level (e.g. if your PRTG web server is not publicly accessible), you can disable HTTPS for the web server. Without encryption web pages are delivered faster and the internal web server is not so heavily stressed. Change this setting in the PRTG Server Administrator program.


4. Sensor Type and Monitoring Considerations


A solid monitoring configuration can considerably improve performance!

Use Longer Intervals

You can reduce monitoring load and data volume by 80 % if you switch the monitoring interval of all of your sensors from 1 minute to 5 minute intervals. Tip: Change the interval in the Root group settings and use PRTG's inheritance mechanism to use this interval for all objects underneath.

Avoid Frequent Auto-Discoveries

Depending on the configuration, an auto-discovery may scan your entire network and therefore produces a certain load. Minimize the number of scheduled auto-discoveries whenever possible. Especially, do not schedule an auto-discovery to run every hour. If possible, set the discovery schedules to Once and run auto-discovery manually when needed.

Use Multiple Probes

Distributing the monitoring load over several probes is always a good idea to relieve your system.

Prefer Basic Network Sensor Types

Basic montitoring types are simple and therefore do not produce much header traffic or system load. These use less system resources than more advanced technologies:

  • SNMP
  • PING
  • HTTP
  • SMTP
  • Port
  • DNS
  • FTP
  • POP3

Use SNMP V1 and V2, But Avoid V3

SNMP V3 doesn't scale well. SNMP V3 involves SSL encryption which creates a lot of CPU load.

Avoid WMI Whenever Possible

Avoid WMI. WMI doesn't scale well. WMI uses DCOM to connect to other computers and has several Mutex/Semaphore issues that affect scaling.

Avoid Network Intensive and the More Complex Sensor Types

Keep usage of the following sensors to a minimum: Sensor Factory, Packet Sniffers, VMware, Email Round Trip, SQL Server, Cloudwatch, QoS, File, Folder, HTTP Full Page, Custom EXE Sensors

Use NetFlow and sFlow Instead of Packet Sniffing

For monitoring of high bandwidth networks, NetFlow and sFlow are far better than Packet Sniffing.

Minimize Use of NetFlow and sFlow Sensors

Depending on the traffic pattern, using NetFlow or sFlow can create a lot of CPU load and data.

Avoid Toplists

Keep usage of Toplists to a minimum. Depending on the traffic pattern and Toplist configuration, this feature can create a lot of CPU load and data.

Avoid Multiple User Accounts and User Groups

Especially in large installations, multiple user accounts and groups can slow down web server performance.

Avoid Defining Many Different Access Rights per Node

Especially in large installations, this could slow down web server performance; depending on the depth of your sensor tree's hierarchy.

Avoid Dependencies

In large installations, avoid configuring many different dependencies. In normal installations, dependencies will be not a problem.

Avoid Schedules

Up to 20 schedules are no problem in any installation.

Avoid Giant Reports

Reports can slow down the system considerably. Only create the smallest reports possible, avoid reports with many sensors over long periods, and run reports at times during the day where nobody uses the webGUI.


5. Exercise Housekeeping


Do what every administrator should do.

Reboot Regularly

We found that servers run more reliable in the long run, when they're rebooted once a month. Also for systems running heavily used Remote Probes consider automatic reboot as often as once per week.

In PRTG 8's Probe Administrator application, you can now set automatic Restart Options.

Defrag Data Drive Regularly

Monitoring software like PRTG constantly writes data and log files to the disk. For large installations the constant flow of data to the disks can reach gigabytes per day. Although PRTG has methods to minimize fragmentation built in, the data on the your disk will fragment heavily (because many large files are written parallel and incremental throughout the day).

Run a defragger once per day: Enable automated defragmentation (built-in for Windows 2008 and Windows 7) or install a defragger that can be run automated daily or weekly.

Tip: We use the Freeware MyDefrag: If you want to use it as well, just create a scheduled task for OptimizeDaily.MyD or OptimizeWeekly.MyD.

Turn Off NTFS Compression for Monitoring Database

During installation, PRTG turns on NTFS compression for the \monitoring database folder. We found that disabling compression actually increases performance for small to medium installations (less data has to be read, less fragmentation). But for large installations this can have the opposite effect.

To disable compression, open the PRTG Server Administrator program, switch to the Core Server tab and un-check the Use NTFS based file compression option next to the database path.

Note: Don't convert already compressed data to uncompressed data! This process causes heavy fragmentation.


6. More Resources and Facts


We conducted a series of tests regarding PRTG's performance on different hardware platforms and operating systems. Please see the following articles from the CEO's blog for more information:


More


Created on Mar 19, 2010 1:40:54 PM by  Dirk Paessler [Paessler Support] (9,613) 3 3

Last change on Oct 13, 2011 10:50:00 AM by  Daniel Zobel [Paessler Support] (21,383) 3 3



Votes:

0

Your Vote:

Up

Down

Created on Apr 6, 2010 7:50:43 AM by  Jens Rupp [Paessler Support] (609) 3 1

Last change on Apr 12, 2010 8:22:52 AM by  Daniel Zobel [Paessler Support] (21,383) 3 3



Votes:

0

Your Vote:

Up

Down

In your article above, you mentioned...

"Keep usage of the following sensors to a minimum: Sensor Factory, Packet Sniffers, VMware, Email Round Trip, SQL Server, Cloudwatch, QoS, File, Folder, HTTP Full Page, Custom EXE Sensors"

We have a fairly large installation of PRTG with over 7,000 sensors. We have quite a few of the above mentioned sensors. It would be helpful to have more solid guidelines on usage of these sensors rather than simply to keep them at a minimum. How many is too many? Can you give us a number or a percentage?

Created on Nov 1, 2010 2:50:34 PM by  Scott K (60) 1 1



Votes:

0

Your Vote:

Up

Down

@Scott K: The sensor types mentioned for "keep to a minimum" are either creating considerable more CPU or considerable more network load so we recommend to avoid using them too often (more than 10-100 sensors per type) in large installations.

On the other hand especially PING and SNMP sensors are creating the smallest footprint of all sensor types, followed by the Internet standard protocols like HTTP, SMTP, POP3, IMAP. WMI uses a lot more cpu cycles, etc.

What we want to say with "keep to a minimum": If you can perform monitoring for specific devices with the less cpu intensive sensors you should do so.

Note: Most of these considerations only come into perspective if you hit any resource limits in the first place.

Created on Nov 2, 2010 7:31:10 AM by  Dirk Paessler [Paessler Support] (9,613) 3 3



Please log in or register to enter your reply.


Disclaimer: The information in the Paessler Knowledge Base comes without warranty of any kind. Use at your own risk. Before applying any instructions please exercise proper system administrator housekeeping. You must make sure that a proper backup of all your data is available.

PRTG
Network Monitor
Intuitive to Use.
Easy to manage.

150.000 administrators have chosen PRTG to monitor their network. Find out how you can reduce cost, increase QoS and ease planning, as well.

Visit
www.paessler.com

What is this?

This knowledgebase contains questions and answers about PRTG Network Monitor and network monitoring in general. You are invited to get involved by asking and answering questions!

Learn more

Top Tags


View all Tags