Just curious, I see that under the probe health, its set at 100 open requests to go into warning. Is that really an issue if you have zero delay low cpu and memory across the board? The console performance is great as well...I have close to 12000 sensors on this probe. Hardware is a new cisco rack mount server.
Just wonder how long I can continue to grow before I need to think about an upgrade. My issue is I need everything to stay in the cluster probe (for the most part, cpu intensive stuff like SOAP and other custom stuff is on remote probes) for DRP reasons and the stability of being able to loose one server and keep everything running.
Article Comments
Thanks for the input.....I was restructuring the sensors\devices in my install, the no netflow sensor allowed in a prtg cluster probe was a large hassle for me, would like to be able just place it under a device but just be a single sensor (not clustered), due to this and the inability to move a sensor from a device to a device, and when you clone a device it doesn't carry over the history has really been a struggle...long story short I either lost netflow or snmp history and have to run one of those side by side for a bit to get some history....that seems really confusing in writing....
long story short I was able to run almost 18000 sensors for a bit in the cluster probe with no hit to performance, cpu, mem, latency or any of the default canned prtg probe alerts, with this many items I still stayed under 50 open requests, with all the delay metrics at "0"....I have since scaled this back down to under 10,000 sensor in the cluster, but was a good test to see how hard I could push the system.
One question, I have the unlimited license, but to do what you stated would result in needing the$ 40,000 dollar (haven't looked in a bit but I know its a lot more) license cost?
Aug, 2015 - Permalink
Dear kube
The unlimited sensor license has no limit regard the sensors, because if you use light sensors only (like pings or SNMP traffic sensors) you can go beyond 10000 while having good performance. An installation with a typical mix of sensors however usually does not scale well beyond 10000 sensors. A cluster usually shows noticeable less performance before you get into the range of 10000 sensors.
Depending on the actual configuration, you can have an 18000 sensor cluster; but this is outside of our specifications.
Please contact sales@paessler.com for any license cost related questions.
Aug, 2015 - Permalink
Dear kube
The open requests are an indicator for the sensor latency. 10000 sensors on a cluster probe indicate a very large installation. If you want to go even further, please consider horizontal scaling, meaning setting up two independent clusters, each running much less than 10000 sensors.
CPU load is not the only indicator for the PRTG performance. I recommend to keep the typical CPU load below 10%. This provides enough headroom for load spikes.
If horizontal scaling is no option for you, please try to increase the sensor scanning intervals. In many cases, an interval of 5 minutes instead 60 seconds provides enough information. This lowers the sensor requests per second drastically and helps the performance.
Aug, 2015 - Permalink