This post is more than 5 years old
115 Posts
0
1344
July 21st, 2011 04:00
nx4 cifs performance
hi guys
i'm migrating data from a netapp and have not moved everything as yet and i was check ing the CIFS performance on what has been moved and is now currently in use.
the CIFS op/sec are quite high at times and i was wondering can someone explain what the ops/sec really represents
i have attached a performance graph - i'm worried if i moved all the data over the nx4 will not be able to handle it
i'm currently only using 1 CIFS server and both pool are archive pools using sas in one and sata in the other
Any suggestions would be greatly appreciated
Beag
No Events found!
umichklewis_ac7b91
300 Posts
0
July 21st, 2011 13:00
This is where the server_stats command is your friend - the options and output change between DART 5.6, 6.0 and VNX, so be sure to use man server_stats, to see your options.. The CIFS ops/sec you see is the cumulative total of all CIFS operations primatives on your NX4. The primatives include various operations to open files, lock them, change permissions, etc. So, even the act of double-clicking to open a file causes multiple primatives on the backend - your Windows client sends more than one command for the open, which results in more than one CIFS operation to open the file from the NX4.
To see the same output, run the command server_stats server_2 -summary cifs - interval 5 -count 5
This shows you five instances of data for a 5-sec interval.
To see a breakdown of this output, run server_stats server_2 -table cifs -interval 5 -count 5
This shows you five instances of much more data for the same interval.
For more information, take a look at the document Managing Statistics on Celerra. You'll see how you can list the available statistic paths (statpaths) to get much more granular details. Let us know if this helps!
Karl
pharford57
115 Posts
0
July 22nd, 2011 00:00
Hi Karl
many thanks for that the stats command does show a lot of stuff, is there a way to see what the max ops are per DM ?
Would creating a second and third vdm spread out the load if they are created on performance fs's ?
Thanks Again
Regards
Paul
umichklewis_ac7b91
300 Posts
1
July 22nd, 2011 05:00
There have been other threads on the "what's the maximum possible performance per DM" and such. The theoretical limits are published in Powerlink, but it's really hard to know exactly what the maximum performance will be - the combinations of your network, network settings (TCP windows, NAS params, etc.), number of spindles, cache, etc. all have an impact on that.
Test suites like IOMeter can atleast factor in your network. We've used IOMeter on test servers (both virtual and physical) and run I/O tests against our Celerra. I'm pretty confident that the numbers I get from IOMeter are atleast "meaningful estimates" of performance. I've found that tracking my stastistics output over time are much more meaningful.
You might try and run an IOMeter test against your current setup, then add a VDM and test the new setup. It might give you a good indication if extra VDMs will balancing things out.
Karl
pharford57
115 Posts
0
July 22nd, 2011 06:00
Hi Rainer
What is ever easy :-)
I gues its talk to my local EMC contact time, is there any prework i can do the help gather information for a performance analysis ?
Or what would you recommended to check to help improve performance ?
i have to do the NAS performance training at some point soon
Thanks
Paul
Rainer_EMC
4 Operator
•
8.6K Posts
1
July 22nd, 2011 06:00
Evaluating ops and performance isnt that easy
No – it doesn’t matter if or how many VDMs you use