Not logged in.
Quick Search - Contribution
Contribution Details
| Type | Other Publication |
| Scope | Discipline-based scholarship |
| Title | Untersuchung der Skalierbarkeit verschiedener Datenbankmanagementsysteme unter hohen Nutzerzahlen |
| Organization Unit | |
| Authors |
|
| How Published | |
| Date | August 2009 |
| Abstract Text | The work presented here describes measurements of transaction throughput for different database management systems that focus on concurrency control. The measurements were taken for IBM DB2 9.5, PostgreSQL 8.3 and Microsoft SQL Server 2008. During the measurements, the follow- ing parameters were being changed to determine their effect on throughput: the isolation level, the amount of memory available for the database’s buffer pool, the database’s cardinality and the amount of operations per transaction. When trying to relate the measurements’ results to the expec- tations based on theoretical principles, it is found that while some effects show as expected, many phenomena have to be attributed to speci?c implementations of the different database management systems. Expected results like lock thrashing and throughput-limitations due to I/O-performance or the CPU’s processing speed are apparent. Unexpectedly, I/O-performance is a limiting factor not only when small database buffer pools are used, and lock thrashing affects all database man- agement systems in a different way. Furthermore, it is found that all of the used management systems can reach higher throughput numbers at higher isolation levels. DB2 is noted to break con- nections when the database’s buffer pool is chosen too large, while SQL Server does the same when the database’s buffer pool is chosen too small. For PostgresSQL, transaction throughput is reduced whenthe level of concurrency is increased. This happens due to the multi-versioning protocol used by PostgreSQL, which leads to an increase in memory consumption under these conditions. |
| Export |
BibTeX
|