Mount options for a deep-learning workstation

by Lucas Figueiredo   Last Updated July 02, 2018 20:00 PM

So, my university advisor assigned me to setup a deep-learning workstation in our laboratory to do some research. But it's my first contact with linux and system administration and I'm feeling a bit lost in some aspects of the installation. One of they is about mount options for each volume.

I'm using LVM to manage the disks (the server have a 480GB SSD and a 6TB HD), it seems that the SSD have plenty of space for all the software and libraries that we gonna use (CUDA, cuDNN, Keras, Tensorflow, etc), but during the LVM configuration of disk partitions, there is some mount options that I can select, like discard, noatime, nodiratime, relatime, usrquota, grpquota, etc.

I already did plenty of research on stack overflow, quora and etc, but I couldn't find any useful information for my case about which options to select.

The idea for the workstation is that everyone from our lab (about 15 students) could use it for processing deep-learning models inside a virtual environment (python).

Ubuntu Server 16.04

CPU: Intel Xeon E5-2620 2,10GHz

64GB of RAM (DDR4)

SSD 480GB

HD 6TB

GPU: 3x Titan X Ultimate Pascal (2016) 12GB

What I'm asking here is some advice about the mount options for the logical volumes (I only created two - root [360GB] and swap [16GB] - inside the SSD) and some suggestions about system administration of deep-learning workstations, from your experiences with this type of setup.



Related Questions


can´t mount sda partition in rescue-grml

Updated January 24, 2018 12:00 PM

Failed to mount external drive while booting

Updated December 16, 2017 15:00 PM

Can we mount multiple disks as one directory?

Updated October 09, 2017 22:00 PM

making sda1 bigger - and why is it so small

Updated February 21, 2016 02:00 AM

How to recover data from a corrupted ext3 partition?

Updated September 20, 2015 21:00 PM