Version 13 by MR
on May 16, 2012 09:36.

compared with
Current by MR
on May 22, 2012 12:45.

This line was removed.
This word was removed. This word was added.
This line was added.

Changes (1)

View Page History

The bigger the data, the less the influence of the job starting overhead. Very interesting to see that reflected on the own experiments - and to see the amount of the influence on performance.

*The 197 GB experiment described above, is an comparable equivalent* to a task we regularly also perform in a "traditional way" (as a java program running on a 4 core HT server). Usually this task takes around 6 hours and 45 minutes to complete in the traditional way. Compared to the 36 minutes of the above mentioned experiment, that is an excellent value for our "poor hardware cluster".