|
|
|
Forum Newbie
      
Group: General Forum Members
Last Login: Tuesday, May 21, 2013 11:59 PM
Points: 8,
Visits: 155
|
|
|
|
|
|
Grasshopper
      
Group: General Forum Members
Last Login: Tuesday, May 07, 2013 4:26 PM
Points: 20,
Visits: 218
|
|
Hi Aldo, rather than set a growth figure would it not be better to set the size so all space is taken up on the drive. e.g. On 40GB drive set each file to 10GB
Thansk
|
|
|
|
|
Grasshopper
      
Group: General Forum Members
Last Login: Tuesday, May 07, 2013 4:26 PM
Points: 20,
Visits: 218
|
|
Hi Aldo, rather than set a growth figure would it not be better to set the size so all space is taken up on the drive. e.g. On 40GB drive set each file to 10GB
Thansk
|
|
|
|
|
SSCommitted
      
Group: General Forum Members
Last Login: Friday, May 10, 2013 4:07 PM
Points: 1,943,
Visits: 8,227
|
|
|
|
|
|
Ten Centuries
      
Group: General Forum Members
Last Login: Today @ 8:56 PM
Points: 1,060,
Visits: 4,182
|
|
Although you may not need a one-to-one match for TempDB files and CPU cores, if you do think TempDB is causing a bottleneck, I'd certainly recommend increasing the number of TempDB files from 1. On a number of our servers I've set up 2 files for TempDB (some servers have 2 cores, some have many more) and even just moving to 2 files seems to improve performance.
I also agree with the comment of making the TempDB files as large as possible - the last thing you want is a delay to a transaction because TempDB has to autogrow. But I wouldn't suggest using the whole disk, maybe 80% - that way you've got room to grow in an emergency while you work out what caused all the space to be used up!
|
|
|
|
|
SSC-Enthusiastic
      
Group: General Forum Members
Last Login: Friday, May 17, 2013 6:54 AM
Points: 183,
Visits: 237
|
|
This one definitely depends on your situation. Most of the time I break the TempDB into only four or eight separate equally sized files with no auto grow. This has been a good general practice for my environment, but each server can be different depending on the type of data on them so testing is the best option. Also in the bulk of the tests I did a couple years ago I found no significant gains in splitting the TempDB files between different Luns/Spindles.
|
|
|
|
|
Grasshopper
      
Group: General Forum Members
Last Login: Saturday, March 16, 2013 12:46 PM
Points: 14,
Visits: 41
|
|
Hello all.
Are those objections true for virtual servers? May be then situation it is more complex since IO is "virtual"?
|
|
|
|
|
Valued Member
      
Group: General Forum Members
Last Login: Wednesday, May 08, 2013 2:50 PM
Points: 57,
Visits: 317
|
|
jts_2003 (7/19/2010) I also agree with the comment of making the TempDB files as large as possible - the last thing you want is a delay to a transaction because TempDB has to autogrow. But I wouldn't suggest using the whole disk, maybe 80% - that way you've got room to grow in an emergency while you work out what caused all the space to be used up!
There's something funky about this logic. You're not increasing space in an emergency; you're artificially limiting space to begin with, and adding the overhead of an autogrow in an emergency. If you think a query might pass that 80% mark, wouldn't it be better to dedicate 100% of the disk to begin with?
|
|
|
|
|
Forum Newbie
      
Group: General Forum Members
Last Login: Sunday, May 01, 2011 5:12 PM
Points: 5,
Visits: 144
|
|
What is the point of having multiple files ON THE SAME DRIVE c:?
|
|
|
|
|
Old Hand
      
Group: General Forum Members
Last Login: Tuesday, May 21, 2013 1:34 PM
Points: 385,
Visits: 611
|
|
obulay-657245 (7/19/2010) What is the point of having multiple files ON THE SAME DRIVE c:?
This is straight from Paul Randal's blog that another user posted a link to abover
PS To address a comment that came in - no, the extra files don't *have* to be on separate storage. If all you're seeing it PAGELATCH contention, separate storage makes no difference as the contention is on in-memory pages. For PAGEIOLATCH waits, you most likely will need to use separate storage, but not necessarily - it may be that you need to move tempdb itself to different storage from other databases rather than just adding more tempdb data files. Analysis of what's stored where will be necessary to pick the correct path to take.
Good article, but great discussion. I just worked with some Microsoft folks on thier SCOM team for a deployment of thier software and they insisted on 1 TempDB per processor and for it to take up most of the drive 80%. It is good to now know what to look for to see if the recommendations were over kill or if they were valid.
It looks likethe answer is with most things SQL Server ....It Depends
Twitter: @SQLBalls Blog: http://www.SQLBalls.com
|
|
|
|