VMware vSphere

DB is not performing  since moving the DB partitions \ Virtual disks all to the 2nd SCSI controller.  

database drives are all using SCSI1, system disk uses SCSI0

There were some issues prior to this, some performance issues here and there, nothing major, but the application owner had concerns.  Originally the database was using a virtual disk that was also sharing SCSI0 with the OS.   WE changed this so that the OS had its own SCSI controller, the the DB virtual disks were all using SCSI1

Now that the virtual disks used by the DB are all on SCSI 1:x the issues seem to be much worse. 

Storage latency is no where to be seen, the back end is flash.  Total latency is always under 2ms, averages .02, with maybe an occasional spike up to 20 something, but never into the 30ms range.

Disk Queue length in the guest OS is never over 1

I could use some advise on this...

I do see that CPU ready has spiked really high at occassions, the app server up to 2650ms and the DB up to 3600ms... this is high... and since the app is multi-tired I fear that this high ready time is causing the problems. 

the datastores in use are configured for Round robin, the storage is HDS and the DB lun is pinned to the flash tier... 

I've been solving performance problems for years and this one has me stumped....  Can someone lend a hand \ tip?   

Thanks

What DB it is?

Pooracle - drains us of funds

Hello kwg66​,

Did you perchance move it to a new scsi controller of a different type that is less optimal for the VMs workload? (e.g. it was on a Paravirtual controller and you added it to a new standard LSI SAS)

About VMware Paravirtual SCSI Controllers

Old but still relevant:

Which vSCSI controller should I choose for performance? - VMware vSphere Blog

Edit: Please also validate if there are any other differences that could affect performance, like did you take snapshots before making the change and forget to consolidate them.

Bob

WE aren't using the paravirtualized SCSI controller, but I had thought of this, I will need to review this a bit my last impression of that controller was the VM needed to be pumping over 2k IOPS.   I'm not so certain it is, the vCenter is v6 so the perf charts don't count IOPS specfically but I could manually add up the reads and writes hitting the DB.  We are upgrading to 6.7 over the weekend where the perf charts are more accurate in regards to packet loss and have more counters avialble. 

Thanks for replying,  There will be a delay in me getting back to this as I am going to be away travelling till next wed.