I have a production deployment using local SSDs for tempDB. I have 2 SSDs in a RAID1 configuration. I am seeing average reads of 1-2ms but the average writes are showing as 1377ms on all four of my tempdb data files.

Each tempdb data file is 2GB with a 1GB growth setting (They haven't grown since deployment 5 months ago)

The tempdb log is showing average read 67ms and average write 215ms.

The SSDs are Samsung 840 pros.

The following code is what I use to get my stats

SELECT a.database_Id, 
db_name(a.database_id) AS dbname, 
db_file_type =  CASE 
                WHEN a.file_id = 2 THEN 'Log' 
                ELSE 'Data' 
UPPER(SUBSTRING(b.physical_name, 1, 2)) AS disk_location, 
a.io_stall_read_ms / Case When a.num_of_reads = 0 Then 1 Else a.num_of_reads end AvgRead,  
a.io_stall_write_ms / Case When a.num_of_writes = 0 Then 1 Else a.num_of_writes end AvgWrite,  
Cast(Round(((( a.size_on_disk_bytes / 1024 ) / 1024.0 ) / 1024), 2) as float) AS size_on_disk_gb                    

FROM sys.dm_io_virtual_file_stats (NULL, NULL) a 
JOIN sys.master_files b ON a.file_id = b.file_id AND a.database_id = b.database_id 

ORDER BY a.io_stall DESC

Below are the top 5 rows returned

database_Id file_id dbname  name    db_file_type    disk_location   io_stall    AvgRead AvgWrite    size_on_disk_gb
2   1   tempdb  tempdev Data    F:  19782846713 2   1377    2
2   3   tempdb  tempdev2    Data    F:  19782655021 2   1377    2
2   5   tempdb  tempdev4    Data    F:  19782364070 2   1377    2
2   4   tempdb  tempdev3    Data    F:  19782151571 2   1377    2
2   2   tempdb  templog Log F:  378829065   67  215 1

So my tmepdb files on SSDs are the slowest drives I have. Anything I should be looking at from a configuration/infrastructure point of view? I am currently studying the applications usage of tempdb and any memory spills but I'm not seeing anything terrible.

  • Do you have any benchmark to compare to? Is the RAID itself performing fast enough for you? Any IO tools show you different result that you see from the DMV?
    – DenisT
    Commented Jun 13, 2014 at 13:07
  • 1
    You could use SQLio to test the performance of the SSDs in a pattern similar to SQL Server.
    – Hannah Vernon
    Commented Jun 13, 2014 at 18:10
  • 1
    RAID controller or software? If controller, is write cache enabled? If so, disable and test again. Does the controller support TRIM? Commented Jun 13, 2014 at 19:03
  • how many processors you have ? what is allocation size on this F drive ? what is showing activity monitor for tempdb related activity ? You have OLAP or OLTP system ? Commented Jun 17, 2014 at 18:53
  • So my sysadmin restarted the server after applying updates and so far the stats coming from SQL Server look good. It may have been one bad process that was pushing the average write time up. I'll monitor and try catch the offending process if or when it happens. Commented Jun 26, 2014 at 7:46

1 Answer 1


We cracked this a while back by replacing the RAID controller on the server. The disks and server configuration were fine but it appears that the RAID controller couldn't deal with the IO.

We are now in the good place of reads ~ 2ms and Writes at <= 5ms

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.