On Aug 28, 2012, at 9:07 AM, Emily Lynema wrote:
> I find this conversation interesting, mostly because the "why do it"
> reasons given parallel so closely what we are working on at NC State in our
> new library building. Except it doesn't have anything to do with
> Our emphasis is on taking expensive visualization and high performance
> computing capacity and making it available to students all across our
> campus. Some would ask why we are building massive visualization walls and
> working on creating a cloud computing environment where anyone can request
> temporary access to high performance computing in order to build "stuff" to
> render on the visualization walls. And it's just the same as the reason
> given for doing makerspaces in academic libraries: while faculty on fancy
> grant projects have access to high performance computing nodes, nowhere on
> campus is this kind of computing and visualization openly available for
> undergraduate students to creatively use.
> It's neat to see the different directions we go with the same underlying
And in that regard, (high performance computing), I heard an interesting
story from someone who I think was from JHU Physics dept. a year or so
Basically, all of the professors were building their own personal beowulf
clusters (getting the money as either part of their condition on hire, or
using grant money to buy them) which caused a number of problems:
1. They weren't experts, so it'd take them a while to set up.
2. They typically didn't secure them properly, so they'd get hacked,
and they had to take them down, and often didn't get them back up
for many months, up to a year from original purchase 'til it was
finally running at full tilt.
(ie, it had already depreciated by a year)
3. So many clusters were built, that it overloaded the electrical
in the building, and the whole building lost power.
So there really are some benefits to having a centralized cluster that
the faculty can submit jobs to, rather than all of the little ones.
The visualization stuff may be even more useful, as they're quite
uncommon. Besides some of the 'hiperwall' and 'cave' systems, there
was a project from one of the Harvard libraries on using a Microsoft
Surface (the table, not the yet-to-be-released table) for working with
huge images (telescope data, hi-res scans, etc.)