Chapter 7. Optimization

• Disk seeks. It takes time for the disk to find a piece of data. With modern disks, the mean time for this is usually lower than 10ms, so we can in theory do about 100 seeks a second. This time improves slowly with new disks and is very hard to optimize for a single table. The way to optimize seek time is to distribute the data onto more than one disk.

• Disk reading and writing. When the disk is at the correct position, we need to read the data. With modern disks, one disk delivers at least 10–20MB/s throughput. This is easier to optimize than seeks because you can read in parallel from multiple disks.

• CPU cycles. When we have the data in main memory, we need to process it to get our result. Having small tables compared to the amount of memory is the most common limiting factor. But with small tables, speed is usually not the problem.

• Memory bandwidth. When the CPU needs more data than can fit in the CPU cache, main memory bandwidth becomes a bottleneck. This is an uncommon bottleneck for most systems, but one to be aware of.

When using the MyISAM storage engine, MySQL uses extremely fast table locking that allows multiple readers or a single writer. The biggest problem with this storage engine occurs when you have a steady stream of mixed updates and slow selects on a single table. If this is a problem for certain tables, you can use another storage engine for them. See Chapter 14, Storage Engines and Table Types.

MySQL can work with both transactional and non-transactional tables. To make it easier to work smoothly with non-transactional tables (which cannot roll back if something goes wrong), MySQL has the following rules. Note that these rules apply only when not running in strict SQL mode or if you use the IGNORE specifier for INSERT or UPDATE.

To change the preceding behaviors, you can enable stricter data handling by setting the server SQL mode appropriately. For more information about data handling, see Section 1.9.6, “How MySQL Deals with Constraints”, Section 5.2.5, “The Server SQL Mode”, and Section 13.2.4, “INSERT Syntax”.

Because all SQL servers implement different parts of standard SQL, it takes work to write portable database applications. It is very easy to achieve portability for very simple selects and inserts, but becomes more difficult the more capabilities you require. If you want an application that is fast with many database systems, it becomes even more difficult.

All database systems have some weak points. That is, they have different design compromises that lead to different behavior.

To make a complex application portable, you need to determine which SQL servers it must work with, and then determine what features those servers support. You can use the MySQL crash-me program to find functions, types, and limits that you can use with a selection of database servers. crash-me does not check for every possible feature, but it is still reasonably comprehensive, performing about 450 tests. An example of the type of information crash-me can provide is that you should not use column names that are longer than 18 characters if you want to be able to use Informix or DB2.

The crash-me program and the MySQL benchmarks are all very database independent. By taking a look at how they are written, you can get a feeling for what you must do to make your own applications database independent. The programs can be found in the sql-bench directory of MySQL source distributions. They are written in Perl and use the DBI database interface. Use of DBI in itself solves part of the portability problem because it provides database-independent access methods. See Section 7.1.4, “The MySQL Benchmark Suite”.

If you strive for database independence, you need to get a good feeling for each SQL server's bottlenecks. For example, MySQL is very fast in retrieving and updating rows for MyISAM tables, but has a problem in mixing slow readers and writers on the same table. Oracle, on the other hand, has a big problem when you try to access rows that you have recently updated (until they are flushed to disk). Transactional database systems in general are not very good at generating summary tables from log tables, because in this case row locking is almost useless.

To make your application really database independent, you should define an easily extendable interface through which you manipulate your data. For example, C++ is available on most systems, so it makes sense to use a C++ class-based interface to the databases.

If you use some feature that is specific to a given database system (such as the REPLACE statement, which is specific to MySQL), you should implement the same feature for other SQL servers by coding an alternative method. Although the alternative might be slower, it enables the other servers to perform the same tasks.

With MySQL, you can use the /*! */ syntax to add MySQL-specific keywords to a statement. The code inside /* */ is treated as a comment (and ignored) by most other SQL servers. For information about writing comments, see Section 9.4, “Comment Syntax”.

If high performance is more important than exactness, as for some Web applications, it is possible to create an application layer that caches all results to give you even higher performance. By letting old results expire after a while, you can keep the cache reasonably fresh. This provides a method to handle high load spikes, in which case you can dynamically increase the cache size and set the expiration timeout higher until things get back to normal.

In this case, the table creation information should contain information about the initial cache size and how often the table should normally be refreshed.

An attractive alternative to implementing an application cache is to use the MySQL query cache. By enabling the query cache, the server handles the details of determining whether a query result can be reused. This simplifies your application. See Section 5.14, “The MySQL Query Cache”.

This section describes an early application for MySQL.

During MySQL initial development, the features of MySQL were made to fit our largest customer, which handled data warehousing for a couple of the largest retailers in Sweden.

From all stores, we got weekly summaries of all bonus card transactions, and were expected to provide useful information for the store owners to help them find how their advertising campaigns were affecting their own customers.

The volume of data was quite huge (about seven million summary transactions per month), and we had data for 4–10 years that we needed to present to the users. We got weekly requests from our customers, who wanted instant access to new reports from this data.

We solved this problem by storing all information per month in compressed “transaction tables.” We had a set of simple macros that generated summary tables grouped by different criteria (product group, customer id, store, and so on) from the tables in which the transactions were stored. The reports were Web pages that were dynamically generated by a small Perl script. This script parsed a Web page, executed the SQL statements in it, and inserted the results. We would have used PHP or mod_perl instead, but they were not available at the time.

For graphical data, we wrote a simple tool in C that could process SQL query results and produce GIF images based on those results. This tool also was dynamically executed from the Perl script that parses the Web pages.

In most cases, a new report could be created simply by copying an existing script and modifying the SQL query that it used. In some cases, we needed to add more columns to an existing summary table or generate a new one. This also was quite simple because we kept all transaction-storage tables on disk. (This amounted to about 50GB of transaction tables and 200GB of other customer data.)

We also let our customers access the summary tables directly with ODBC so that the advanced users could experiment with the data themselves.

This system worked well and we had no problems handling the data with quite modest Sun Ultra SPARCstation hardware (2×200MHz). Eventually the system was migrated to Linux.

This benchmark suite is meant to tell any user what operations a given SQL implementation performs well or poorly. You can get a good idea for how the benchmarks work by looking at the code and results in the sql-bench directory in any MySQL source distribution.

Note that this benchmark is single-threaded, so it measures the minimum time for the operations performed. We plan to add multi-threaded tests to the benchmark suite in the future.

To use the benchmark suite, the following requirements must be satisfied:

After you obtain a MySQL source distribution, you can find the benchmark suite located in its sql-bench directory. To run the benchmark tests, build MySQL, and then change location into the sql-bench directory and execute the run-all-tests script:

shell> cd sql-bench
shell> perl run-all-tests --server=server_name

shell> perl run-all-tests --help