Large Scale Installation & Dependencies / Notifcations

Dear sparish

Dependency checks take CPU time and they need to create a read-lock on the device tree, blocking write accesses onto that tree during the check.

Please clarify, how large is your setup? How many sensors do you use and on what machine (physical or virtual, number of logical CPU cores, RAM)?

Hi,

We're talking about 7000 sensors, soon to be much more. Currently we mainly use PRTG to ping most of the equipment in our remote sites using http sensors / scripts, but I am going to start rolling out more advanced monitoring for places it may be useful (SNMP). So I would expect this number to be many times larger

We have 2 Physical Core XL1 servers, which are both:-

Intel Xeon E5 series CPU (12 CPU cores, 24CPU threads) 32GB RAM 4 x 250GB Solid State Disks RAID 10 array ~ 470GB usable space Windows 2012 R2

The load will be spread across both, how would you manage notifications without dependencies in large deployments?

Dear sparish

Before I can give a useful reply, please clarify "load will be spread across both" – do you plan to use two independent core server, or a cluster/failover solution?

Hi,

Two independent core servers, we'll use the Enterprise console to be able to see them both at the same time without having to go to two URLs.

For us, it made the most sense as we have a local and remote backup for both machines, so in the case of hardware failure we'll be up and running in a few hours.

Hi,

Related to the above, I am looking at using dependencies to reduce the barrage of notications after a device goes offline as we're going to have several sensors per device. As a test I have a router, hypervisor and VM being monitored. Each is it's own device under one group with multiple sensors per device and following best practices, each device has a simple sensor such as ping / port that I have set as the master object with a 60 sec scanning interval (the others have a higher scanning interval).

This means that per device I will only get one notification if there is an error, which is great but I can't seem to set the device's dependencies once it has a sensor set as the master object. If my router dies, I only want one notification saying my router is dead not 3 saying my router, hypervisor and VM are dead, as I am aware the other 2 will be dead if my router is missing. In this example it is only saving 2 notifications being sent but in a network with 72 access points & 6 switches sat behind the VM, we're talking about 81 notifications when that router drops. How do I get a device to depend on another device, when it has a sensor set as the master object?

Dear sparish

You could create a group, move the devices there, and set the master object for that group to a specific ping sensor.

Hi,

Thanks for the reply. It seems a little odd, that if on a device I have a master sensor in order to pause the others, that device can no longer depend on another device. In my above scenario, the only way to get around that would be to create a group for every device that has other devices depending on it? That would mean a group for every device and then another for each group of AP's that depend on a different switch. It's not deal breaking, it's just extra work and I was wondering why? I can understand it may be difficult to visualise what exactly I mean, so if it's easier I can send a support email instead with a diagram.

Dear sparish

To keep the dependency path clear, we don't allow multiple dependencies. Overall we think that the benefits outweigh the cons.

In principle, you could disable the "master object for parent" setting and set the same ping as master for all three devices which should then be paused. But this way you don't get individual device pausing.