Upgrade planning

[Predrag Punosevac]

Dear Autonians,

It has been brought to my attention that the CUDA upgrade from 10.3 to 11.0
didn't work as expected for some people. That is very unfortunate in
the lieu of incoming deadlines. The servers are like living beings. They
are constantly metamorphosing even if I don't do anything. Sitting for too
long on the existing software versions has its advantage that the system is
working fully or mostly for 5-6 months but it has an unintended consequence
of forcing us to do updates at the least desirable times before conference.

I was very reluctant to upgrade from 10.3 to 11.0 but I had to after a
multitude of reports about broken TensorFlow.

Due to the fact that my training is in pure mathematics, I lack the basic
awareness of major conferences and deadlines in the field of Machine
Learning. Typically this is not a problem during the noral times as people
just walk to my office and tell me to keep my hands on the equipment for
the next 2 weeks. These are not normal times so I will have to do a bit
better job. I am trying to create a calendar of block out weeks when the
changes to their computing infrastructure will be kept to the minimum. I am
thinking of 2 solid weeks before any major conference or due deadline.

Could you please kindly help Ben and I to do this by filling out the poll
he created

https://docs.google.com/document/d/13GeOx5IM7zxS2IJjPoJweHFKKw8-PYLjL8Z9vG0G8i0/edit?usp=sharing

Best,
Predrag
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/autonlab-users/attachments/20200901/0137d305/attachment.html>