Few things deserve the attention of a long rant as much as unneccessarily complicated syntaxes. When you want to achieve something that is clearly defined and supported, but you have to look up the the syntax. PIVOT and UNPIVOT are examples of such features, and in this case, I’ll even show you a more well-performing alternative.
For windowed functions, SQL Server introduces two new operators in the execution plan; Segment and Sequence Project. If you’ve tried looking them up in the documentation, you’ll know that it’s not exactly perfectly obvious how they work. Here’s my stab at clarifying what they actually do.
In the two previous parts of this series, we’ve looked at how parallelism works, how you can control it, and how it affects your query (and server) performance in different environments. In this, the third part, we’re going to take a more technical look at how the different Parallelism operators work.
Continuing on last week’s post on parallelism, here’s part two, where we take a closer look at when parallel plans are considered and what you can do to either force or prevent a query from running parallel as well as things you want to avoid if you’re trying to achieve a parallel query plan.
The subject of parallel execution of SQL Server queries is at times somewhat shrouded in mystery and uncertainty. Since the concept of parallel execution is such a significant (and indispensable tool) for performance tuning, it’s good to have a fair idea of how it works. In this first post in a series on parallelization, I’m going to try to sort out the apples from the pears, or the serial from the parallel if you will.
There are a handful of wonderful software packages out there that I wouldn’t want to work without, including the free, open-source virtualization platform VirtualBox, which was acquired by Sun a few years ago, which in turn was bought by Oracle.
The other day, I was setting up a new virtual machine to run SQL Server 2014 with memory-optimized tables, which incidentally is one of the great reasons to update to 2014. Memory-optimized tables are tables that are stored in the RAM memory of the server. Some of the great advantages of working with “in-memory OLTP”, as it’s also known, include a greatly reduced number of latches and locks (accomplished with a form of row-versioning), which allows for a much larger number of concurrent users to work with the same data. With a few limitations, working with memory-optimized tables is transparent, so you’re using regular T-SQL DML commands.
Turns out, however, that SQL Server didn’t want to run memory-optimized code right out of the box on my VM. Instead, I got this:
Msg 41342, Level 15, State 2, Line 3 The model of the processor on the system does not support creating MEMORY_OPTIMIZED=ON. This error typically occurs with older processors. See SQL Server Books Online for information on supported models.
This message stems from the fact that the processor needs to support the CMPXCHG16B command (I have no idea what that does), which is available on pretty much any modern 64-bit processor. My physical processor is fairly new, so the issue is obviously with the VM software. In this case, it was just a matter of enabling a setting in VirtualBox, which can be done with the following command:
VBoxManage setextradata "Your VM" VBoxInternal/CPUM/CMPXCHG16B 1
I have no idea why this feature isn’t turned on by default, but once you’ve enabled it, it works like a charm. So, expect to see more blog posts about memory-optimized tables in the near future. :)
Here’s a practical script that I built to view how much space each database object takes up in the database. The script also show you information on how many rows a table (or indexed view) contains, as well as if it’s partitioned and/or compressed.
Every user session in SQL Server comes with a number of “flags”, or settings, that you can alter to modify the way SQL Server behaves in different aspects. Some are fairly straight-forward, which others are old legacy options or even completely deprecated and have no actual use.
Today, we’re going to be looking at a kind of poor-mans’s-partitioning, using a view to union records from multiple tables. We’ll also take a look at when you would want this type of solution, some benefits and drawbacks, as well as ways to make things go faster.
If you’re working with data warehousing or reporting, you’ll recognize this problem as a recurring headache whenever you’re designing an ETL process for fact tables: If you want to completely reload all the rows of a fact table, you would typically start by emptying (or truncating) the fact table, and then load new data into it. But during the loading process, depending on what your job does, there won’t be any data in the table, or worse, it will be half-filled and incorrect. Worst-case: If your ETL job crashes, the table will remain empty. Now, if your ETL job takes an hour to run, that’s a problem.