Thanks everyone for the feedback regarding the slowness problem. We are releasing a new version that will address the slowness problem. It will require an upgrade from the prior version. We're doing everything we can to get the fix available within a week. Also, we apologize to those of you affected by the problem.
Carter Shanklin
That's great news!
I appreciate the way you have asked the opinion of the community ánd have acted accordingly.
Our update is posted, http://vmware.com/go/powershell
That great news! I think.
I have just tested this new build, but the results are a bit disappointing:
Build 103777 ("old"):
Measure-Command { Get-VM }
Average of 10.5 seconds for 210 vms.
Build 113525 ("new"):
Measure-Command { Get-VM }
Average of 10.5 seconds for 210 vms.
EDIT: wait! these were the results for the beta:
200 vms
beta: 11 seconds
GA: 51 seconds
Huh?!
Message was edited by: hugopeeters
We don't have any VMs with vmxnet NICs in our test environment.
They all have "flexible" NICs.
The response time with the new build is 35-40% slower.
Can someone confirm this ?
My conclusion: if you don't have the vmxnet problem stick with the 103777 build.
Blog: lucd.info Twitter: @LucD22 Co-author PowerCLI Reference
Hi Luc,
I'm very surprised by this, can you post the times it takes to retrieve your VMs (using measure-command) before and after?
Also, does it tend to take the same amount of time each time you call get-vm?
Thanks, Carter
I don't see any improvement either. With 1.0 it would take 25 seconds, after the new build is installed it still takes 25 seconds to run get-vm for about 200 VMs.
Hi Hugo and John,
The performance fix was not an overall speedup, it was to address a bug that caused queries to be really slow (up to 10 minutes in some cases) if you had VmxNet adapters on your virtual machines.
We're considering some techniques we can use for overall performance improvement but this is a bigger issue and will take longer to solve. In the meantime we welcome your feedback on the overall performance and how it affects the things you're trying to do with the toolkit.
Carter
You may want to check out some of the operations used in Alan's contest entry for benchmarking. It took a good 20 min to run this against my VI.
Author of the upcoming book: Managing VMware Infrastructure with PowerShell
Co-Host, PowerScripting Podcast (http://powerscripting.net)
Hi Carter,
I double-checked my figures but I'm afraid my observations were confirmed.
I used a Get-VM for a single, powered-off guest.
Note that the NIC of this guest is defined as "flexible", not a "vmxnet".
I used this short test program to perform the benchmark and write the results to a file.
$report =@() for($i=0; $i -lt 100; $i++){ $report += (Measure-Command {Get-VM PC11}).TotalSeconds } $report | Set-Content "bench.txt"
The test was performed with build 103777 and 113525.
The results are in the attached CSV file.
The average for both tests gave:
- build 103777: 0.258254448
- build 113525: 0.828925418
Which is in fact worst than the 30-40% I mentioned earlier.
Blog: lucd.info Twitter: @LucD22 Co-author PowerCLI Reference