2015-10-26 03:46 AM
I was wondering if you can actually count on the write IOPS reported by the DFM Performance Manager from a 7mode Metrocluster ?
I know that if you do a "sysstat -u 1" in the commandline, you will have to halfe all the write operations shown, because the mirrored plex is also "counted" here.
I'm pretty sure that a statit also shows this, so I was wondering if the data the DFM reports in it's Performance Manager also has to be halfed in order to get the real write IOPs ?
This is beause we want to move a workload from a Metrocluster to a HA system (non Metrocluster) and has to estimate the IOPS.
I just want to be cirtan that we are not calculating with the double ammount of writes ;-)
Solved! SEE THE SOLUTION
2015-10-26 06:06 AM
Can you explain with an example?
May i know which counter is refered here?
Can you please provide the screenshot of the NMC performance manger, the counter which you are refering to ?
2015-10-26 07:27 AM
From the commandline you can do a: "sysstat -u 1" which gives you this output:
CPU Total Net kB/s Disk kB/s Tape kB/s Cache Cache CP CP Disk
ops/s in out read write read write age hit time ty util
92% 5803 41196 60773 116914 96790 0 0 15s 94% 70% H 58%
95% 9631 42175 103122 98741 33761 0 0 16s 95% 100% : 61%
79% 6908 33563 109377 99579 3058 0 0 16s 94% 23% : 71%
93% 6714 20196 83347 122120 66092 0 0 16s 95% 95% H 58%
78% 6987 38888 108745 122491 46359 0 0 17s 95% 100% : 35%
84% 6833 51326 120124 102073 21804 0 0 14s 94% 55% H 38%
Here you can see the Disk write kB/s which in a MetroCluster shows both data written to the local plex and the remote plex, so you have to devide this in half in order to get the "real" numer of writes. This is also reflexted in the Write IOPS the controller does.
I would like to know if this is also the case in the DFM Performance Manager couters like "Disk Throughput":
Or maybe also here in the Volume ops ?
Or does DFM realize that this is a MetroCluster with a Mirrored Plex and device the IOPS ?
2015-10-28 01:43 AM - edited 2015-10-28 01:44 AM
With SyncMirror (which MetroCluster uses) the writes are doubled at the raid layer. So counters that are higher (like volume, protocol, system IOPs) will display 1x, while counters that are lower (raid, storage, disk, system disk KB ) will have 2x.
I hope that helps!
Storage Architect, NetApp EMEA (and author of Harvest)
Blog: It all begins with data