Improve SQL Performance

Recently we are assigned a task to improve the performance of a couple of SQL queries that run for 15 seconds and 30 seconds. At the beginning we were not so sure whether the performance can be improved anymore, given that there were about 80 million data in the database. But I have to admit that SQL Server is really powerful as eventually we are able to reduce the time down to 2 seconds and 6 seconds!

During the process, I have had a few interesting findings about SQL Server and learned a lot about troubleshooting SQL Server performance issues. Here I want to share a few tips on optimizing SQL statements I learned from the experience.

check Index first

We sometimes get ahead of ourselves and jump directly into analyzing SQL queries, but the answer to the performance issue may just lie on the indexing itself, so always check whether necessary indexes are presented. A good way to find out what indexes you need to is to use Dynamic Management Views to monitor system health and come up necessary indexes. With appropriate indexes in place, performance increase will be substantial.

Clustered vs. Non-clustered index

There are tons of articles about Clustered and Non-clustered index and their benefits. Knowing the difference between clustered and non-clustered index helps troubleshooting performance issue for sure. Here is a nice article explaining Clustered and Non-clustered indexes.

Here I would like to share a metaphor I like which helped me understand Clustered and Non-clustered index – Clustered index is like the page numbers of a book where the logic order of the pages is the same as the physical order of them in the book, e.g. page 1 is at the first and page 100 is the last page for a 100 pages book. On the other hand, Non-clustered index is like the Index page of a book where the logic order of the terms on the index page is not the same as their actual locations in the book. For instance, the word Apple is listed at the beginning of the index page but it may firstly appear on the last page of the book.

Analyze, trial and error

To really find out why a certain query is running slow, more information is needed.

VIEW Execution plan and set Statistics io

Execution plan can be quite helpful to figure out which parts of a complex query are taking time and how SQL server optimized the query. Though it does not provide a solution directly, it does provide some hints or directions on troubleshooting performance issues. SQL Management Studio or even Visual Studio now provide the ability to view execution plan. A good tool I find quite useful is SQL Sentry Plan Explorer which displays the execution plan diagram with much more information, for example, it displays the number of records processed above each line, which is very convenient.

Another quite handy tool, Statistics IO, displays statistics information for your query.  Set Statistics IO ON and information such as the number of scans will be displayed.

Break nested queries

However, execution plan may not  be very straightforward sometimes, especially for a complex query with many nested sub-queries. The large query will be optimized by SQL Server which makes the execution plan harder to understand. Instead of trying to figure out what SQL Server is doing,  a faster way for me is to just break the large query into smaller ones and test the performance of each of them. It is possible that SQL Server is not picking the best route to run the query and the ‘optimized’ query is actually running much slower, which is exactly what happened in my case. My query should only takes 2 seconds to run whereas the ‘optimized’ query by SQL Server takes 15 seconds.  Why? Optimizing query also takes time and SQL Server may just not find the best optimization in time.

If SQL Server is selecting a slower way to execute the query, there are a couple of workaround:

  1. Use OPTION (FORCE ORDER) at the end of the query to force the order of join to be the same as it is in the query statement.
  2. Break nested sub-queries into temporary tables. This way you can force SQL Server to materialize sub-queries.
Trial and Error

Another quick and easy approach is just try different ways of writing the same query. Swap IN with JOIN, change INNER JOIN to LEFT JOIN to EXIST,  or change Non-Unique Non-clustered index to Clustered index. Some changes just take a few seconds to make and they may just work. Once it works, you can research a bit on why or may just choose to forget about it if not in the mood. I happen to know someone (me) who is always not in the mood 🙂

How to write to Windows Azure Website’s application log and more

We struggled with Windows Azure website because we couldn’t write logs. Using Log4Net to write to log file as we used to do, does not work for Azure Website because of the limited file permission. We had to resort to email notification or Virtual Machine when we needed to debug Azure Websites, which was a big headache!

Fortunately, it is all over now. To write to the Application Log of Azure Website, just use System.Diagnostics.Trace name space and use method like TraceInformation, TraceError, TraceWarning to record different levels of log!

Trace.Wrietline("Log Verbose level log");

Trace.TraceInformation ("Log Information level log");

Trace.TraceWarning("Log Warning level log");

Trace.TraceError("Log Error level log");

Then just turn on the Application Logging and select a logging level for that Azure Website.

4-16-2015 5-41-11 PM

With this feature, it becomes much easier to troubleshoot Azure websites. Even better, Microsoft provides this streaming log function from which you can view application logs in REAL-TIME! (New Azure portal only)

4-16-2015 5-29-27 PM

Furthermore, here is something developers will definitely like – this streaming log is also available in Visual Studio, and you can filter the result using Regular Expression! (Latest version Azure SDK is required)

4-16-2015 5-33-57 PM

4-16-2015 5-36-13 PM

Since file logging is supposed to be turned off automatically after 12 hours, if you also want to log into a table storage, not a problem. You can set up a Azure Storage to hold the log.

4-16-2015 5-42-41 PM

Click View settings of the Azure Website in Visual Studio.  In the Log tab, there  will  be a nice table view of the log. I do notice that it uses a lot of memory of the Azure Website. Just something to consider.

4-16-2015 5-43-48 PM

That’s what I know about logging to Azure Websites. Hopefully it is helpful. We use Azure on most of our web applications and I think it is just getting better everyday. Now with the ability to write application log for Azure Website, it just meets all of our needs, but there is still a huge set of exciting features we haven’t used. I look forward to exploring those someday!