Just a general question to the community, how high can a single collector go in terms of number of agents and metrics ?
In the sizing guidelines CA say there is "no known maximum load that is built into the Enterprise Manager software"
With that in mind we were wondering what sizings can be realistaically achieved.
Does anyone want to share examples of their largest numbers and the spec that the collector hardware is running on ?
At our location, our largest collector under full load has -
Version - APM 9.5.2
Linux version 2.6.18-371.8.1.el5
Product Name: ProLiant DL380p Gen8
32 available processors
Storage is SAN.
In 9.0.8 I used to regularly run up against introscope.enterprisemanager.metrics.historical.limit=1200000 on an EM that shared 2 CPUs on a P7 AIX frame.
the recommendations in the CA APM Performance and Sizing Guide were tested in labs. It really depends on the hardware (for collectors primarily IO and CPU/memory for MOM) and load that is made up of many factors that are discussed in the guide. As long as the EM capacity is steadily at ~60% and both Harvest and SmartStor duration are in the green area (see Cookbook - EM HealthCheck v20.pdf) you can increase the load in small steps).
Remember to have at least headroom for your busiest collector to go down (taking your load balancing configuration into account)! Otherwise the snowball effect of a single collector failure will crash you entire cluster
Hey thanks for that Guenter, thats a great document I hadn't seen before.
What we are trying to achieve at our location is to migrate from AIX collectors to LINUX based collectors, and in the process reduce our collector footprint down from the 10 collectors we have on this cluster. But we don't exactly know to what number.
Our aim at the moment is to size the collectors with an upper limit of 1 million metrics handled.
Do you think that is achievable with appropriate specced hardware ?
Yes. That is definitely possible. Look at Example 3 and 4 at recommendations. Linux typically performs a bit better than Windows - and a lot better than AIX. As stated above the most important thing for the collectors is I/O. Check harvest and SmartStor durations on your current collectors. On a healthy Enterprise Manager, the harvest duration is less than 3500 milliseconds most of the time. Use local disks or SAN. Have dedicated disks for SmartStor (data directory).
P.S. The CA APM Performance and Sizing Guide has been part of standard APM documentation for years
Thanks Guenter - yeah I have used the sizing guidelines documents for years, it was the cookbook I hadn't seen before. I should have been a bity clearer there !