With the growing trend of customers leveraging Windows Server 2012 Hyper V in the field, there is an expected corresponding trend in inquiries around data collection and identifying the performance requirements of Hyper V deployed on CSV targets.
At the time of this writing, the usual EMC SE reporting does not accurately report on the disk performance characteristics of CSV targets. The way these engagements typically go is:
- PERFCOLLECT is collected from the Hyper V host
- The data is then uploaded for analysis
- The report returned returns data only for the LOGICAL DISK counters identified in the PERFMON file (namely, the C: drive)
By golly, where is all that IO that I KNOW is there? The quick response is that there IS performance data in the collect. We are presently working to streamline the rendering that we are used to getting.
In the interim....
I am processing the BLG file collected via perfcollect using PAL (http://pal.codeplex.com/) – which will give IOPS and response times. If you’d like to try, install PAL and open it. It’s wizard based. Point it at the PERFMON file and select the desired BLG file.
This will at least get us IOPS, read/write splits and the performance characteristics of the host in a pretty easy-to-read format. Probably something wrong with the way Mitrends is parsing the data – at least PAL will give us the state of the hosts.
Download and install PAL, and run it.
In the Counter Log page, browse to the BLG file in the PERFMON subfolder of the PERFCOLLECT data.
Then click Next.
Then select the Threshold File that you want to use. It has profiles for most of the Microsoft applications that PERFMON grabs counters for – in this case you are looking at a Microsoft Windows Server 2012 Hyper V host – so select that one.
If you keep hitting Next through the rest of the Wizard it will start processing the BLG and generate an HTML report with alerts and data in a report. It works pretty nicely – and each area of the report explains what the counter MEANS in the context of the application selected in the Threshold File that you selected.
Physical Disk\Disk Transfers Per Sec
This is probably the best counter (for now) to estimate IOPS. Check queue depths to see if the workload is being met – if there are sustained queue depths there may be a choke point – either with HBA queue settings or the spindles are not meeting the load.
Physical Disks/Bytes per Sec will give you the throughput.
There is a fantastic reference for Hyper V Performance Guidelines available for download from Microsoft. Storage gets interesting at around page 65...
Please remember that IOPS are CUMULATIVE for the cluster - capacity is NOT!
How are YOU doing it today for your Hyper V environment? Feedback and discussion is welcome - I would love to hear how the rest of the EMC Microsoft Community (customers, partners, and colleagues) is answering the call!
Cunningham's Law states: The best way to get the right answer on the internet is not by asking a question. It's to post the wrong answer!