esxtop statistics
esxtop collects an abundance of statistics regarding all aspects of how your ESXi host and VMs are performing. Although it is possible to view metrics in regards to CPU, memory, and networking as well, I've only outlined the storage-related metrics here. The thresholds listed here are simply suggested values, thresholds that I've seen used in many whitepapers, VMware documentation, and resources in the past. There are many different reasons why a threshold could be met, so these are certainly not hard numbers.
Have a look at the following table:
|
Statistic |
Description |
Threshold |
|---|---|---|
|
CMDS/s |
Number of commands issued per second. |
varies |
|
READS/s |
Number of read commands issued per second. |
varies |
|
WRITES/s |
Number of write commands issued per second. |
varies |
|
MBREAD/s |
Megabytes read per second. |
varies |
|
MBWRTN/s |
Megabytes written per second. |
varies |
|
DAVG/cmd |
Latency observed by the device driver—roundtrip latency from HBA to storage array. Sustained thresholds usually indicate a performance issue with the underlying storage. |
25 |
|
KAVG/cmd |
Latency observed inside the VMkernel. Value should always be very low if not 0 unless queuing is observed. |
1 |
|
QAVG/cmd |
Latency observed inside the queue. This is part of KAVG/cmd. Sustained values indicate an issue with queuing or queue depth. |
1 |
|
GAVG/cmd |
Round trip latency as observed by the guest OS. Normally a total of DAVG, KAVG, and QAVG. |
25 |
|
AQLEN |
The storage adapter queue length—maximum number of active commands the adapter is configured for. |
n/a |
|
LQLEN |
The LUN queue depth—maximum number of active commands the LUN can have. |
n/a |
|
WQLEN |
The world queue depth—maximum number of active commands the world can contain. |
n/a |
|
ACTV |
The number of commands that are currently active within the VMkernel. |
varies |
|
QUED |
The number of commands that are currently queued in the VMkernel waiting for processing. Sustained thresholds may indicate a need to increase queue depth or an issue with the underlying storage array. |
1 |
|
%USD |
The percentage of queue depth used by active commands. Normally sits close to |
1 |
|
LOAD |
The total number of active and queued commands as compared to queue depth. Should always be 0 unless queuing is occurring. |
1 |
|
ABRTS/s |
The number of commands that have been aborted per second. Normally indicates that the underlying storage is unable to meet the demands of your workloads. |
1 |
|
RESETS/s |
The number of commands reset per second. |
1 |
|
RESV/s |
The number of SCSI reservations issued per second. |
n/a |
|
CONS/s |
The number of SCSI reservations conflicts occurring per second. Sustained high values could indicate that actions need to be taken to balance the metadata heavy operations. |
20 |
