New computing node

[predragp at andrew.cmu.edu]

Dear Autonians,

About a month ago I built very powerful computing node for Foxconn project
(foxconn.int.autonlab.org). I just got permission from Dr. Dubrawski and
Dr. Schneider to make it available to all members of Auton Lab with the
understanding that priority when it comes to use of this server will be
given to members of Foxconn team. You can access this computing node with
your regular LDAP credentials just like any other machine in the lab.

Hardware specs:

2xIntel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz 8 core each (16 hyper-threads)

RAM 384 GB of the latest and fastest

storage space 6 TB effective (most of it will be used by foxconn but you
scratch directories are there too)

NFS is mounted with new and improved options (I am allowing script
execution directly from NFS). Still be realistic about the speed which is
in real terms 80-90 MB/s due to 1 Gigabit network limitation. Theoretical
speed of NFS on this network is about 120 MB/s. To give you an idea how
fast/slow this is I will remind you that new SSD drives including the one
I use to install OS on this server have speed of over 600 MB/s. RAID 6
with XFS peeks around 350 MB/s. That means that using local file system to
do data science is at least 4 time faster than using NFS.


Finally some great software news for this server. It runs Red Hat 7.1
which is compiled with gcc-4.8.3. Feels much faster than 6.6. I installed
latest MATLAB R2015a, R, python-2.7.5 is now default system Python. Usual
suspects are there too: Pandas, scikit-learn, numpy, scipy, ipython, blas,
atlas, boost, gsl, and the whole slue of other software you are accustomed
to.


Enjoy,
Predrag