What are the performance implications of placing an APM Cluster on one box for Production (colocation)?

Document ID : KB000030098
Last Modified Date : 14/02/2018
Show Technical Document Details

Description:

Often customers place all their EMs on one server. (Colocation) The APM Sizing and Performance Guide discusses these cases:

 

1) Running multiple Collectors on a single computer.

2) Running multiple MOMs on a single computer.

3) Running a MOM with multiple collectors on a single computer.

4) Colocation with the APM Database and some combination of above on a single computer.

5) Colocation with an EEM and some combination of above on a single computer.

 

What are the tradeoffs of doing this?

 

Solution:

If not following the hardware/software recommendations, the following performance conditions may occur:

 

* Unanticipated resource contention conflicts (Network connections, Smartstor disk thrashing, CPU/Heap usage etc.)

* Slow performance

* System(s) crashing

 

Should the system crash, you could stand to lose all your Smartstor Data, APM DB and Traces DB if they all reside on the same server.  This is very bad for Data Recovery purposes as you have a single point of failure, which can significantly increase downtime, for your entire cluster or multiple clusters depending on your configuration.  Ideally you would want only one EM per server because if you lose one server, you can still run with the cluster. (i.e. If the MOM server is lost and you have a failover MOM configured on another server, then you will have very little downtime.)  You can keep your cluster running and reporting properly while trying to get the original MOM operational.

Without a failover MOM or using one EM per server, we cannot get data back if the disk on the server is hosed.  Each EM contains their own Smartstor DB and traces DB.  The MOM nor any other EM does not make a copy of an EM's Smartstor DB, Traces DB and APM DB.