get_stat memory counters

[Guv]

I have got the below script which reports on memory usage of VM's via using the get-stat command.   In our virtual centre out statistci levels are set to 1 for each interval duration, so we can only use basic counters.   The main counters for memory I have used are mem.usage.average and mem.consumed.average.  The script is below :

$businessDays = "Monday","Tuesday","Wednesday","Thursday","Friday"
$dayStart = New-Object DateTime(1,1,1,7,30,0)
$dayEnd = New-Object DateTime(1,1,1,17,30,0)
get-cluster "EO Cluster 0" |Get-VM | where {$_.PowerState -eq "PoweredOn" -and $_.MemoryMB -ge 4096} | `
Select name,MemoryMB,
  @{N="Average Memory Usage Percentage";E={[math]::Round((get-stat -entity $_ -Start ((Get-Date).AddDays(-7)) -Finish (Get-Date) -stat mem.usage.average |Where-object {$businessDays -contains $_.Timestamp.DayOfWeek -and $_.timestamp.timeofday -gt $daystart.timeofday -and $_.timestamp.timeofday -lt $dayend.timeofday} | Measure-Object -Property Value -Average).Average)}},

@{N="Average Memory Usage MB";E={[math]::Round((get-stat -entity $_ -Start ((Get-Date).AddDays(-7)) -Finish (Get-Date) -stat mem.consumed.average | Where-object {$businessDays -contains $_.Timestamp.DayOfWeek -and $_.timestamp.timeofday -gt $daystart.timeofday -and $_.timestamp.timeofday -lt $dayend.timeofday} | Measure-Object -Property Value -Average).Average / 1KB)}}

This script gets values only from VM's which have more than 4GB of memory.

The issue I have is understanding the counter values and in particular the counters.  For one of the VM's, i get the below values for the last 7 days :

Name: VM Name

MemoryMB: 4096

Avg Memory Usage : 24      (mem.usage.average)

Avg Memory Usage MB : 4096   (mem.consumed.average)

or

Name: VM Name

MemoryMB: 6144

Avg Memory Usage : 8   (mem.usage.average)

Avg Memory Usage MB : 5879 (mem.consumed.average)

I just wanted to find out what the mem.consumed.average value shows, as from the above value the average mem consumed is quite high same as the memory allocated but the percentage used is quite low.  So which value do we follow.  Does the consumed value shows that how much of the guest memory has been touched but not used.   Can anyone please advise, as I need to explain these reports to management.

[LucD]

In the PerformanceManager page for memory metrics you will find the definitions for both metrics.

The mem.usage.average metric expresses a percentage showing the actively used memory over the configured memory. The mem.active.average metric shows the actual amount in MB. These are the pages, according to the VMkernel, that are recently touched by the OS in the guest.

And the mem.consumed.average metric show the amount of memory that is reserved by the guest OS, but that might not be actually in use.

In your first example, the guest OS has allocated 4 GB of memory.

But the OS is only actively using (touching) 24% (or 983 MB) of that allocated memory, as seen in recent samples by the VMkernel.

[vAdmin]

Thanks Luc for your blog article about the performance metric in http://www.lucd.info/2011/07/08/powercli-vsphere-statistics-part-5-rollup-types/ and providing us the script, however I wonder why I could not add two more metrics in the following script:

$metrics = "cpu.usage.average","cpu.ready.summation","cpu.swapwait.summation","mem.active.average"
$entities = Get-VM
$start = (Get-Date).AddMinutes(-5)
 
$report = Get-Stat -Entity $entities -Stat $metrics -Realtime -Start $start | `
Group-Object -Property EntityId,Timestamp | %{
  New-Object PSObject -Property @{
    Name = $_.Group[0].Entity.Name
    Time = $_.Group[0].Timestamp
    CpuAvg = ($_.Group | where {$_.MetricId -eq "cpu.usage.average"}).Value
CpuRdy = ($_.Group | where {$_.MetricId -eq "cpu.ready.summation"}).Value
    CpuSwpWait = ($_.Group | where {$_.MetricId -eq "cpu.swapwait.summation"}).Value
    MemActAvg = ($_.Group | where {$_.MetricId -eq "mem.active.average"}).Value
    
  }
}
$report | Sort-Object Name,Time | Export-Csv "C:\VMstats.csv" -NoTypeInformation -UseCulture

[LucD]

These metrics return values for each instance and an aggregate (average over the instances).

That means the Where-clause returns more than 1 object and hence the result is an array and the Value property is not there.

The solution is to take the aggregate value for these 2 metrics.

Something like this

$metrics = "cpu.usage.average","cpu.ready.summation","cpu.swapwait.summation","mem.active.average" 
$entities = Get-VM $start = (Get-Date).AddMinutes(-5)

$report = Get-Stat -Entity $entities -Stat $metrics -Realtime -Start $start | 
Group-Object -Property EntityId,Timestamp | %{
    New-Object PSObject -Property @{
        Name = $_.Group[0].Entity.Name
        Time = $_.Group[0].Timestamp
        CpuAvg = ($_.Group | where {$_.MetricId -eq "cpu.usage.average"}).Value
        CpuRdy = ($_.Group | where {$_.MetricId -eq "cpu.ready.summation" -and $_.Instance -eq ""}).Value
        CpuSwpWait = ($_.Group | where {$_.MetricId -eq "cpu.swapwait.summation" -and $_.Instance -eq ""}).Value
        MemActAvg = ($_.Group | where {$_.MetricId -eq "mem.active.average"}).Value

    }
}
$report | Sort-Object Name,Time | Export-Csv "C:\VMstats.csv" -NoTypeInformation -UseCulture 

[vAdmin]

Ah ok, now I understood, Thanks Luc for the reply.

I appreciate your efforts in sharing and helping the community here.

Cheers.

[shishir08]

Memory Consumed is the actual amount of Machine Memory that has physical memory mapped to it.

Memory Consumed = Memory Granted - Shared Pages.

So it may happen that the sharing of pages between the VMs may be less thus Memory consumed value is very high.

Memory size presented to the guest can categorised in two way.

1) Allocated memory-Memory assigned to applications.

        Furthe it can be categorised in two ways:

               a) Active Memory-Allocated memory recently accessed/used by application.

               b) Idle Memory- Allocated memory recently not accessed/used by application

2) Free memory-Memory not assigned.

I thinks this will clarify your doubts.

HTH

shishir

[vAdmin]

Shishir, Thanks for the explanation.

What is "Shared Pages" actually use for ? is this shared memory for use by the hypervisor (TPS) ?

[shishir08]

Transparent page sharing is something that continuously keeps on running on the ESX host.TPS do not runs at VM level.So suppose you have powered on 4 windows VMs on the machine then what TPS will do is that it will start identifying the pages which are shared across VMs and will only keep one copy of it,that way we will save lot of memory.These are some of the way through which we achieve memory overcommitment.