September 5, 2019 at 12:11 pm
Hello,
I have a question about a certain DMV.
Recently we started a test of our new storage environment. We want to see if its suitable to work for our SQL Server.
The server we are using is virtual and placed on an ESX cluster. We use RedGat SQL Monitor to monitor the server and we are able to use the metrics from the ESX host in RedGate SQL server too. The storage engineers are ablt to see the performance on the storage device.
Until now we have run several test. test that will have a lot of inserts and selects.
If we look at the Avg read/write /sec in SQL Monitor then we see there is hardly any large latency spikes. Most of the time it stays below 5ms. This sounds good.
Now another way to monitor IO latency is to use the DMV sys.dm_io_virtual_file_stats.
There are several articles with an explanation.
If we use this DMW, we see a latency of 219ms on the Tempdb disk. This is not good. But strangly enough the perfomon counters don't show any latency.
Can someone explain this? Who is telling the truth?
Kick
September 5, 2019 at 12:53 pm
well, direct hardware is always telling the truth, VM's and anything above might lie to you.
have another test and compare the following: Does your TempDB during 215 ms latency show any QD (Queue Depth) > 1 on the volume? If so, does the host show QD > 1 aswell? If not, there is your issue. In that case the VM would stack QD > 1 but isn't able / allowed to pass QD > 1 to the storage. You might need to either change the SCSI Adapter for the VM or even upgrade the ESX Host (the latter one VMware would tell you once you open a support ticket).
Viewing 2 posts - 1 through 1 (of 1 total)
You must be logged in to reply to this topic. Login to reply