by Jungh K
For this week’s blog assignment, I chose an article titled, “Hash join in commercial database with Flash memory SSD”, by Sang-Shin Park and Sang-Won Lee, from Sungkyunkwan University in South Korea. There are different types of join operations, such as nested-loop, hash, and sort-merge, in a relational database management system. Hash join operation, being the fastest, is used the most frequently. However, when dealing with large amount of data, memory overflow can occur. In order to prevent this, hash tables are migrated to a magnetic disk, which slows down the process significantly. The article experiments with both magnetic disk and flash based SSD to compare calculated costs by query optimizer. The results show that flash memory SSD (SLC) is about 9 to 20 times faster than 7200RPM magnetic disk. Also, SSD has a consistent correlation with the hash processing times. On the other hand, optimizer frequently miscalculates the expected costs in the magnetic disk. Overall, using SSD as a temporary storage medium in a database greatly reduces performance degradation.
In the class lecture, we learned about physical database design in a relational database management system. In order to optimize performance, data are constantly indexed by differently methods on the background. Especially with Big Data, hash joins are the integral process in the current database management system.
I personally started to use SSD about 2-3 years ago and instantly fall in love with it. With flash memory price declines each year, I believe that it should replace traditional magnetic disk drives as a primary physical drive soon. Even though Windows Server 2008 R2 supports up to 2TB of RAM, the price of 2TB of RAM will be ridiculously expensive and therefore SSD should be a good alternative to maintain the optimal system performance.
Sang-Shin Park; Sang-Won Lee; , “Hash join in commercial database with flash memory SSD,” Computer Science and Information Technology (ICCSIT), 2010 3rd IEEE International Conference on , vol.4, no., pp.265-268, 9-11 July 2010