Each collection interval has a default collection level that determines the amount of data gathered and which counters are available for display in the charts. Collection levels are also referred to as statistics levels.

Table 1. Statistics Levels
Level Metrics Best Practice
Level 1
  • Cluster Services (VMware Distributed Resource Scheduler) – all metrics
  • CPU – cpuentitlement, totalmhz, usage (average), usagemhz
  • Disk – capacity, maxTotalLatency, provisioned, unshared, usage (average), used
  • Memory – consumed, mementitlement, overhead, swapinRate, swapoutRate, swapused, totalmb, usage (average), vmmemctl (balloon)
  • Network – usage (average), IPv6
  • System – heartbeat, uptime
  • Virtual Machine Operations – numChangeDS, numChangeHost, numChangeHostDS

Use for long-term performance monitoring when device statistics are not required.

Level 1 is the default Collection Level for all Collection Intervals.

Level 2
  • Level 1 metrics
  • CPU – idle, reservedCapacity
  • Disk – All metrics, excluding numberRead and numberWrite.
  • Memory – All metrics, excluding memUsed and maximum and minimum rollup values.
  • Virtual Machine Operations – All metrics
Use for long-term performance monitoring when device statistics are not required but you want to monitor more than the basic statistics.
Level 3
  • Level 1 and Level 2 metrics
  • Metrics for all counters, excluding minimum and maximum rollup values.
  • Device metrics

Use for short-term performance monitoring after encountering problems or when device statistics are required.

Level 4 All metrics supported by the vCenter Server, including minimum and maximum rollup values.

Use for short-term performance monitoring after encountering problems or when device statistics are required.

Note:

When the statistics levels, level 3 or level 4 are used beyond the default value, it may cause one particular process, vpxd, to sustain memory growth, if it cannot save the statistics information to the database as quickly as required. If the usage limit of these statistics levels is not monitored closely, it may cause vpxd to grow out of memory and eventually crash.

So, in case the administrator decides to elevate any of these levels, it is necessary for the administrator to monitor the size of the vpxd process to make sure that is not growing boundlessly after the change.