Thursday, September 26, 2013

Performance Tuning In LA!

Hello Dear Reader!  This year Pragmatic Works has brought our Performance Tuning Workshop to Chicago, Boston, and Atlanta.  Due to the success we've added one more stop before the end of the year, Los Angeles!

That's right Pragmatic Works is headed to the City of Angels on December 9th - 10th for two days of Deep Dives and Performance Tuning goodness.

"So Balls", you say "What will you be covering in you're Performance Tuning Workshop?"

Great question Dear Reader!  We start off our journey at the bottom talking about hardware, we discuss the details of CPU's, Memory, and Disks and how their throughput and proper configuration can prevent Hardware Bottlenecks.  We cover installing SQL Server talking about our Pre-Installation Check List and our Post Installation Checklist.

Then we tackle Data files, Round Robin and Proportional Fill, Partitioning, and Compression.  Next we dive into the Query Optimizer to discuss how the Relational Engine works and how it works with the Storage Engine, along the way we'll stop to look at Execution Plans and Plan cache.  From there we dive into the world of Indexes, covering what makes the best Clustered Index, Non-Clustered Indexes, Columnstore Indexes, Index Maintenance, Duplicate Indexes, Reverse Indexes, Unused Indexes, and Missing Indexes.

And that's just Day 1.  We've still got Locking, Blocking, Latches, Wait Stats, Baselines, Alerting, and Extended Events before we are through.

As a member of the class you will get all of our slide decks and Demo's to follow along and keep for yourself.  Lunch is also provided each day as well as coffee and donuts or bagels and muffins.

For all the details Click Here to go to the course outline.  Let's not forget that you get access to myself, and my co-Presenter for those two days.  You'll have a lot of questions, and during brakes and after the class you'll have access to us for Q & A.

Plus you'll get to network with other great folks from the SQL Community and your local SQL Server User Groups.


All this and I still haven't announced my co-Presenter for this yet!  We'll save that for another blog.

"So Balls", you say "Why are you being a tease?"

Yes, Dear Reader.  Shamelessly it is a tease, but I promise it will be worth it.

And now for one last shameless plug.  The price.  Two days of community, access to a top notch as yet to be announced speaker..... and me....., breakfast, lunch, and probably some swag all for $399.  Right now we also have an Early Bird rate until November 9th, it is only $299.


We are already actively planning our courses for next year Dear Reader.  As soon as I have the finalized schedule I will be posting it.  But right now we have NINE 2 day Workshops we are planning, FIVE 4 1/2 day virtual classes on Performance Tuning, and TWO 5 day week long Bootcamps that will be staffed by a couple SQL MVP's and an MCM.

We will be all over the US, hopefully in a city near you, and presented Virtually throughout the Globe.

Good stuff is coming and I'm very excited to be part of it!  As always Dear Reader, Thanks for stopping by.



Tuesday, September 24, 2013

Can You Compress a Temp Table?

 Hello Dear Reader!  We are finishing up the final day of the Performance Tuning Workshop here in Atlanta and I got an interesting question on Compression from Tim Radney (@tradney | Blog).

The question: Can you compress a temp table? Just a quick blog to get the answer out there while Gareth Swanepoel (@GarethSwan | Blog)  teaches the class about Extended Events. 

My guess was yes.  Temp Tables can have statistics, Clustered and Non-Clustered Indexes, while they only exist in the session they are created, I would be they could be compressed.  If you would actually want to compress them is a different discussion, but let’s prove this out.


Here’s a quick demo to show you can do this.  So first up we will create our Temp Table specifying with Data_Compression=ROW.

This will create our temp table #myTable1, we will then insert 15000.

if exists(select name from tempdb.sys.tables where name like '#myTable1%')
     drop table #mytable1
create table #myTable1(
              myid int identity(1,1) primary key clustered
              ,mychar1 char(500) default 'a'
              ,mychar2 char(3000) default 'b'
              ) with (data_compression=row)
declare @i int
set @i=0
     set @i=@i+1
     insert into #myTable1
     default values

Now let’s use DBCC IND to view the pages associated with our table, and DBCC Page to Validate that our data is compressed.

dbcc ind(tempdb, '#myTable1', 1)

dbcc traceon(3604)
dbcc page('tempdb', 1, 376,3)

Looking at the output of DBCC Page I can see that the CD array for my compressed data is present near the header.  Row compression is indeed on.

Now let’s rebuild this using page compression on a rebuild operation using sp_spaceused to measure the size of the table.

And it is now Page Compressed.  Thanks for the question Tim!  And as always Dear Reader Thank you for stopping by.



Tuesday, September 17, 2013

SQL 2014 SSMS is Killing my C:

I could hear SQL saying to my C: Drive "Why You Little...."
 Hello Dear Reader! 

The Blog about SQL Saturday 232 coming shortly, but first I needed to blog about an error that I’m getting.  I presented this weekend on SQL 2014 a First Look at What’s New.  One of the demo’s I did was using Buffer Pool Extensions. 

To confirm with what I’m seeing I tested this on 2 VM’s and one physical Instance.

I created VM1 on my SSD.  I allocated 4 processors and 8 GB of RAM for my VM.  I created 3 drives for data files M: , log files L: , backup files W:, Buffer Pool Extensions S: .  I then Installed CTP1 using Oracle’s Virtual Box.  Max Memory set to 4096 MB.

I also confirmed these results by creating a Hyper-V VM on our Host server in the Pragmatic Works Lab.  The guest has 4 processors and 8 GB of RAM.  Max Memory set to 4096 MB.  No BPE was used on this VM.

Physical Instance
I also tested this on my laptop on a SQL 2012 SP1 CU2 instance.  I have 4 cores and 16 GB of RAM.  Max Memory set to 6144 MB. 

I then used WinDirStat to confirm the results.  During my SQL Saturday presentation my VM unexpectedly ran out of space on the C drive.  This was a little confusing as I had specifically set up my SQL instance to not be installed on the C drive.

“So Balls”, you say, “What was taking up all that space on the C drive?”

Great Question Dear Reader!  That’s want I wanted to know as well.


This is the VM.  As you can see I didn’t go with a huge C drive, only 25 GB.  The only drive I didn’t mention above was my T drive that I used to store a database that I put a Clustered Columnstore Index on.

I’ve got a script that I need to blog on called MakeAdventureWorksDW_big, but I don’t have that typed up just yet.  I use that script to make my FactInternetSales_big table.  I used my friend and cohort Robert Cain’s (@ArcaneCode | Blog) script Make Adventure Works Modern for Today to make my AdventureWorks2013 Database.

My table has 42.9 Million rows in it and is a little over 8 GB.  My test for PBE was pretty simple.  I created a 15 GB BPE.  The instance has 8 GB.  My max memory for the instance was set to 4 GB.  Then I raised the max memory to 19 GB.  I ran sys.dm_os_buffer_descriptors using the is_in_bpool_extension to find the data pages as they were allocated.

SELECT DB_NAME(database_id) AS [Database Name]
, case is_in_bpool_extension
     when 1 then 'Stored in BPE'
     when 0 then 'Stored in non-BPE cache'
  end as BufferLocation
,COUNT(*) * 8/1024.0 AS [Cached Size (MB)]
FROM sys.dm_os_buffer_descriptors
WHERE database_id > 4
AND database_id <> 32767
GROUP BY DB_NAME(database_id), is_in_bpool_extension

I then ran the following query to load up as much data as possible.

Use AdventureWorksDW2013
set transaction isolation level read uncommitted
set nocount on
select * from [dbo].[FactInternetSales_big]

It is just a simple select statement.  Rerunning my buffers query I could see things were loading up nicely.

My plan was to leave this query running so I could load all the data I needed to into my BPE.  That is when the C: ran out of space and the VM froze on me.  

In order to catch this bug I then ran WinDirStat.  Looking under my C:\Users\SQLBALLS.PW\AppData\Local\Temp\1 folder I found this.

Hmmm…. right now not that bad still plenty of free space, but here’s this temp file.  Let’s fast forward to 3 minutes in.  We’re now at 2 GB.

At 5 minutes we are at 3 GB.

Looking at my C Drive, I’ve now dropped from 10 GB of free space to a little over 7 GB.

The result set is large enough that this will continue until my C drive runs out of space.  If I try to delete the file I get a nice error letting me know that the file is locked by SQL Server Management Studio.

15 minutes in my C drive is looking pretty scary.  We are now in the red.  I kill the query so I can keep looking at my VM.  The C drive doesn’t free up.  I know from experience I could now delete the temp file because it is no longer in use.

I asked the question on twitter and Jason Kyle(@JasonNKyle) replied with a suggestion that maybe I was looking at a Swap file. 

As I understand them a swap file allows an operating system to use hard disk space to simulate extra memory when the system runs low on memory.  The way this behaves is that the OS swaps a section of RAM that an idle program is using onto the hard disk to free up memory for the program you are using.  Then when you switch back to the other program the OS trades out the memory for bytes on the HD.

I don’t think this is a swap file though.  I could be wrong. 

When I open one of the smaller temp files in notepad, the query results from my my query re-run so it is a much smaller result set.  This is what I see.

Not really useful, except that I realized the first data that I highlighted happened to be my sales order number from my query.

As I continue to mine through this I see that this is truly my query result set.

If I run this same query on my laptop.  16 GB of RAM, 6 GB Max Memory, after a restart with only around 1 GB in the buffers, I get the exact same results.  At first I thought this was an issue with Buffer Pool Extensions.  Now it looks like this is just the way queries return on SSMS.

Granted you normally don’t run million row transactions in SSMS.  Your application runs them.  So more digging to go, but I wanted to get the blog out to attract comments and help show what I was seeing.

As always Dear Reader, Thanks for stopping by.