Unsolved
This post is more than 5 years old
45 Posts
0
859
VDI Performance
Ive been asked by my manager to check the performance of our VNX5300 to give him a warm feeling regarding our VDI implementation. The trouble is I'm not sure what metrics to use. We are using shared pools for the datastores, and are using FAST Cache for these, so looking at spindle level perfromance isnt going to help. Looking at the system cache and the it's sitting happily between the high and low watermarks, so what can i use to convince him the array is working fine?
SushantGulati
55 Posts
0
February 19th, 2013 04:00
Hi,
Would it be possible for you to post NAR files for the peak load times or probably one full day?
However, to start you still have lots of parameters available. Also, these are just for starters, you can always go down deeper.
a) SP : Utilization
Port Queue Full Counts
SP Cache Dirty Pages %
SP Write Cache Flushes/sec
b) Disks which are part of the storage pool:
Utilization
Queue Length
Average Busy Queue Length
Service and Response Times
c) Lun :
Utilization
Queue Length
Average Busy Queue Length
Service and Response Times
Total Throughput
I think these should give you a fair idea of how your array is behaving. However, if you can still post the NAR files, it will be more helpful.
Tr1gger1
45 Posts
0
February 19th, 2013 05:00
Many thanks for the reply. So are there any limits i could apply to these metrics to see if its under any stress? What sort of queue length is acceptable, what would be an unnacceptable response time etc etc?
Regards
SushantGulati
55 Posts
0
February 19th, 2013 21:00
Sure.
Utilization for disks should be below 70%.
SP Utilization should also be below 70%, in the 50%-70%.
SP write cache dirty pages % should fall in between the HWM and LWM set in the System properties. Its typically set to 80 and 60.
SP Port Queue full counts and Flushes/sec should be as low as possible.
Service time FC 15k disks is in the range of 3-5 secs. Value higher than that seen consistently should be investigated.
Response times are a relative thing. But as a rule of thumb anything above 20ms should be confirmed whether its an issue..
etc.etc.
Can you please post the NAR file in this discussion so that me and other members can take a look at it?
Tr1gger1
45 Posts
0
February 20th, 2013 04:00
Please find attached NAR files as requested.Thanks.
2 Attachments
CKM00120300968_SPA_2013-02-20_11-42-57_2f317b_data.zip
CKM00120300968_SPB_2013-02-20_11-48-45_2f3154_data.zip
Rainer_EMC
8.6K Posts
0
February 21st, 2013 13:00
Or go and download the VNX monitoring and reporting
comes with a 30 day eval license and doesn't need Analyzer
Joshua_Fawcett
35 Posts
0
February 21st, 2013 13:00
Those are actually the SP Collects and do not contain a lot of the performance metrics. To get the nar files you will need to start data logging and let it run for a few hours before a nar/naz file will be generated. To get to these you will need to go to System>Statistics for Block in Unisphere. The "Performance Data Logging" option in the top left will allow you start data logging and let it run for a while to generate the files. That same page has a Retrieve Archive Option in the top right quadrant that will allow you to retrieve these files.
Do you have the Analyzer license?
Do you have a reason to believe performance is not good? Or is it more that you manager just wants to see how things are performing?
I personally like to look a lot at the repsonse times on the LUNs involved. If they are at or under 20 ms, things are usually running pretty smoothly. The metrics Sushant provided above are quite useful as well.
SushantGulati
55 Posts
0
February 22nd, 2013 02:00
I've given the Menu path to download the NAR files.
System > Monitoring and Alerts > Statistics for Block > Archive Management > Retrieve Archive