**disable_resource_pool_perf_metrics. This is the major one for collection time improvements. This disables all metrics of type PERF (visible in the GUI) on resource pools. The vCenter has to go to the database to get these metrics, so any IO limitations with the database will be exposed. This also disables perf metrics on Virtual Apps (a special type of resource pool that we track separately).
**show_networks. If NetworkStatus or alarms/events on networks (in the tree @ ?My Resource ->?My Datacenters?->My Datacenter->network) aren't used, this can be set to "no", for minor memory savings and collection time improvement. This is very minor.
**perf_request_batch_size. Defaulted to 100, this controls how many entities we request information for at once. Extremely low numbers (sub 10) produce awful collection times but can be useful for debugging. Some systems seem to get slightly faster collection times by tweaking this parameter. Unfortunately, we have not determined any pattern for when they benefit from increasing it and when they benefit from decreasing it. It seems to be a matter of playing with it.
**disable_host_system_perf_metrics This is another minor optimization. If no metrics on the System node (or its children) are being used, this can be set to "yes" to save a little bit of memory and a little bit of time in the collection.
**disable_array_metrics Minor optimization. This disables collection of host virtual switches, host services and VM guest disks. Additionally, if any status info metrics were turned on in the mondef . cfg , it disables them as well.
Finally, increasing the available memory via the startup options can help. If they're constantly running near the memory limit, they should increase their initial memory allocation (-Xms) and strongly consider raising their max memory allocation (-Xmx).