As part of my capacity reports I collect the most meaningful metrics for our capacity reports. There are thresholds involved as well. So the elaborate question is as follows
I heard CPU Ready and I/O Wait Time are good capacity indicators in ESX environments? What are the best practice recommendations and thresholds for these values? Also we are strictly looking at capacity at the cluster /host level as opposed to the VM b/c how can you predict capacity on VM's when their allocations are dynamic. Therefore look at the overall hit the host/clusters take in terms of metric thresholds. I still report the top 10 busiest vm's to monitor growth of the vm but the ultimate indicator is in respect to the hosts that they are on....like do we need to add another host to the cluster to support an additional application or individual vm growth - Is this a good approach?
I need some best practice verbiage too. Here is an example of CPU and memory stats and how it looks in my reports ......would you be able to provide this information for CPU Ready and I/O Wait
CPU: % Utilization
BASELINES: Total CPU utilization starts to near capacity thresholds at approximately 75%. At sustained periods of greater than 80% CPU utilization may be a sign that the processor is exceeding capacity and may be the bottleneck.
MEMORY: Pages/Second
BASELINES: Memory Pages/Second tracks page faults requiring data to be read from or written to disk. Application data must wait for this data to be read, so high values may impact application response times. Best practices suggest that sustained values exceeding 50 to 150 pages per paging disk (depending on the speed of the disk) may require further investigation as it may point to a disk bottleneck. Due to the large variation in page activity by applications, the threshold has been set at 200 pages per second. These numbers can usually be alleviated by adding more RAM or potential configuration changes.
