3Par Fibre Channel Virtual Volume shows as ISCSI

[mwkirk]

We are having an issue with a 3Par host where we have a 12 TB VV presented as an RDM to a Windows 2012 Vmware VM.

This drive appears to have poor performance compared with other similar Drives and this same VM has another RDM that appears to perform normally.

The ONLY thing I see different is when we look at this LUN in devices in Vmware as a 3PARdata ISCSI Disk but has a valid FC identifier of naa.60002ac0000000000000000900003d6e. Also looking in some of the performance counters in Vmware this Lun shows high Physical Device Latency at times.

The disk showing up as ISCSI may be a red herring but at this point I am just looking at one of these things is not like the others.

Anyone ever seen anything like this?

[hdtvguy]

What SCSI controller are you using in the VM? Paravirtual SCSI would be the recommended controller. What are you using to gauge performance for the RDM other than vmware perf counters, have you run any Windows perfmon counters? Are the NTFS cluster sizes the same for both RDMs?

[Richard Siemers]

Can you confirm the lun settings in the 3par side to check into the FC vs ISCSI export?

[mwkirk]

It was presented as Fibre-Channel but had been previously presented as ISCSI so according to HP that is why it picked up the ISCSI disk identifier and it should not really affect anything and is just a name.

[Architect]

maybe slightly offtopic, but If you spot less then optimal performance issues, please look at the growth sizes and step sizes of your CPGs.
- growth sizes for FC and NL should be 32GB*the number of nodepairs (so assuming a 4-node 7400, it should be 64GB). For SSD it should be 8GB*the number of nodepairs (16GB for a 4-node 3PAR)
- default step sizes depend on Raid level used, the set-size and drive type. Often I advice to remove the -ss value all together from the CPG configuration, to force the 3PAR to use the defaults (which are always right). Of course there are very heavily tuned CPGs possible where you could need to vary the stepsize based on the underlying application, but usually the default is best.

e.g. via CLI do the following to show the CPG configuration:

Code: Select all

showcpg -sdg

output would be something like:

Code: Select all

Id Name                  Domain            Warn   Limit  Grow Args
Some_FC_Raid5_CPG        SomeDomain           -       - 16384 -t r5 -ha cage -ssz 4 -ss 32 -ch first -p -devtype FC  

in this example both the growth size is to small 16GB instead of 64GB), and the step size is wrong (32KB vs 128KB). setting the right value is done like this:

Code: Select all

setcpg -sdgs 65536 -t r5 -ha cage -ssz 4 -ss 128 -ch first -p -devtype FC Some_FC_Raid5_CPG
 

The system will require a "tunesys" afterwards to adjust the right stepsizes for all previously created LDs with the "wrong" specifications. Please suspend AO during a tunesys to prevent nasty issues.

Below a reference with the right stepsizes per drivetype / raid / setsize:

For HDDs (both NL and FC):

Code: Select all

R1:     256k
R5:     128k
R6 4+2: 128k
R6 6+2:  64k
R6 8+2:  64k
R6 10+2: 64k
R6 14+2: 32k

For SSD's:

Code: Select all

R0:      32k
R1:      32k
R5:      64k
R6 6+2:  64k
R6 8+2:  64k
R6 10+2: 64k
R6 14+2: 32k