Caching done wrong

[Predrag Punosevac]

Dear Autonians,

One of the servers (gpu6) is about to crash due to unskilful use of
python cashing mechanism. When done right cashing can improve speeds and
efficiency of computing. When is done wrong cashes can quickly fill up
/tmp directory. On the default Red Hat /tmp is the part of /root slice
which is typically of size 50-60GB. Our typical installation requires
about 30GB of space ever since I moved MATLAB to a separate slice.
Please make sure that you are cashing things in your scratch directory
and that you are deleting them frequently. If you don't know how to do
it please ask for help.

Best,
Predrag