DFM 3.7 New Performance reports meaning

[Dion_Rivera9]

If you would please find the definition of latency and how it is calculated for the following Reports  –

Volume Performance Summary
Volume NAS Performance Summary
Volume SAN  Performance Summary

I am seeing very high latency numbers that do not match with perfstats.  Could it possibly be the total ops and total latency for that volume that is displayed?

[adaikkap]

Hi David,

The latency for a protocol is calculated as average for of all the ops. The very high latency number is due to a bug in PA 3.7.1 which is fixed in 3.8 scheduled to be released in May 2009.

Please refer to Burt 327609 for more details

Regards

adai

[nagesha]

Hi David,
Volume Performance Summary report shows value of  'avg_latency' as exposed by Counter Manger, this value should match with perfstat.
However other latency fields (nfs_latency, cifs_latency and san_latency) available with rest of the reports may show wrong values, Burt 327609 is filed and fixed in DFM3.8.  And to answer your query these fields indicate average latency of all the ops,  so it would be average of read, write and other latencies of corresponding protocols.

P.S. DFM3.8 is tentatively planned to release around time frame  April/May 09.

Regards,

Nagesh

[jeremyrigefsky]

I am new to Performance adviosor.  I noticed the number for Avg_latency seemed too high and found this posting.  I searched for the "Burt 327609" listed on teh NOW site and found nothing.  can you provide any more detail on the issue?  is there a way to manually calculate the proper avg_latancy?  is just the Total_ops Divided by the Avg_latency being reported?

[nagesha]

Hi,

As far as I know  there are no known issues in PA for  volume:avg_latency metric. Burt327609 is specific to only volume:cifs_latency,nfs_latency,san_latency metrics.  I would suggest you  to file a burt if the issue is specific  to volume:avg_latency.

>> is there a way to manually calculate the proper avg_latancy?

possible, but we have to get the raw values from the filer and then perform necessary math.

(Time t0)            avg_latency =5000000, total_ops =2000000
(Time t0+x)        avg_latency=5006000 ,total_ops=2003000

avg_latency =  (5000600 - 5000000)/ (2000020 - 2000000)

alternatively run below cli on filer

jameel*> stats show -i 2 volume:vol6:avg_latency
Instance avg_latency
                  us
    vol6       28.00
    vol6       24.00
    vol6           0
    vol6           0

Regards,

Nagesh