We experienced problems in running Word2Vec on the server despite of sufficient heap and off-heap memory allocations. It was then found out that we had just 1gig given to Tomcat instead of the 2gigs recommended. The problem went away at the time, but now problems with Word2Vec have resurfaced. Does it make sense to try to increase the memory allocation for Tomcat further? What kind of levels are you guys generally running with?
Tomcat memory is not relevant in this case. All jobs are run in the executor and consume memory there. Therefore only the executor memory allocation matters.
Tomcat memory becomes relevant if you have a large workflow repository, run many jobs at the same time, or have a large number of concurrent users.
But the 1g Tomcat memory was indeed identified by the KNIME Support as the bottleneck that caused the workflow with Word2Vec and subsequently the Executor to crash and require a restart by the server admin. And as said, changing it to 2g removed those problems at the time.
I am running the same workflow successfully on my 32gb RAM laptop with xmx 20g and off-heap 16g. The server has 64gb, the Executor has xmx 30g and off-heap 16g and isn’t burdened by other big workflows at the time, especially considering that it doesn’t need to run Outlook, Excel, Tableau etc simultaneously as my laptop.
But form your answer I can see that this is not something people would generally be experiencing, which means we need to continue our investigation beyond the memory settings as to why the server goes down so easily.
Is it possible for you to create a support ticket by mailing us at email@example.com and we can troubleshoot the issue together.
Of course, but I just wanted to hear out from other users here as well, users who must run must bigger applications that we do.