One of the toughest task for a proper Operations Manager capacity planning is the storage subsystem dimensioning. We have good indications in the OpsMgr Design Guide (Operations Manager 2007 R2 Design Guide) and we can even use an excel sizer (Operations Manager 2007 R2 Sizing Helper). Being involved in OpsMgr planning pretty often I started to collect my own real world metrics, in this post I share my findings. Since this is my reference article for disk capacity planning I’ll keep updating it.
To start and have a general understanding of the topic you can read Jonathan’s post http://blogs.technet.com/b/jonathanalmquist/archive/2009/04/06/how-can-i-gauge-operations-manager-database-performance.aspx.
Since we don’t have any percentile measure in OpsMgr Data warehouse I based my conclusions on a 30 days observation period with an hourly aggregation, collecting the maximum average value for every single disk. I feel the maximum of the hourly averages is a good indication of what we really need from the storage subsystem.
I focused my analysis on Windows 2008 R2 since this will be the platform Operations manager 2012 will require for the product infrastructure. The observed environment is managing about 650 agents and extensively uses dedicated gateways. The RMS hasn’t any gateway or agent directly connected, all the gateways are managed by a dedicated Management Server. Obviously your mileage may vary especially on the database side, there are simply too many factors to take into account: number of consoles, console polling frequency, level of alert noise, number of management packs, number and frequency of reporting, …
|Workload||Iops||% Read||% Write||Notes|
|Management Server||200||5||95||I tried to subtract the OS load, since the agent cache is on the system disk|
|Management Server with OS||300||20||80|
|RMS||160||5||95||I tried to subtract the OS load, since the agent cache is on the system disk|
|RMS with OS||260||35||65|
|Gateway 20 agents with OS||200||40||60|
|Gateway 80 agents with OS||270||35||65|
|Gateway 100 agents with OS||300||30||70|
|Gateway 120 agents with OS||270||20||80|
|Gateway 150 agents with OS||390||20||80|
From the above table, with the interesting exception of the average for the gateways in the 120 agents load, we can infer the IOps demand grows linearly and once agents are over 50/60 it’s safe to suppose about 2 IOps per agent for gateway servers.
Hope this can be of help in your capacity planning effort.
This posting is provided "AS IS" with no warranties, and confers no rights.