In the next Spark release (2.0), I’‘m making some fundamental changes to handle this by updating some of the VM parameters for better performance (and better memory handling — fingers crossed). On my local systems, I’'m not able to reproduce this issue, but I know it exists
it should be really easy to reproduce. Start another instance of Spark in the morning and don’'t use it the whole day. In the evening you may want to try to open it and check whether this runs fine or takes some seconds for disc IO.
I cannot reproduce this with 1.1.9.6. Could you get the alpha build and see if this is still a problem for you. The new VM settings in those builds should help out.