University of Florida
High-Performance Computing Center
News
06/11/09: Many new publications
Many new publications have been added to the website of the last couple
of days as mor e researchers respond to the requests sent out. Now, it
is even easier to submit a publication that references the HPC Center
by simply filling out the
Publication Submission form online.
06/02/09:
ICBR and IFAS lead HPC Phase III Expansion
In January 2009, the Phase III expansion of the UF High Performance
Computing (HPC) cluster was completed. The effort to build the coalition
started in early 2007, with planning complete in November of 2007. This
expansion brought the total computing capacity of the HPC cluster to
2,500 cores, which translates to about 10 Tflops (floating point
operations per second).
Full Article
05/12/09: Penguin Powers University of Florida HPC Expansion
04/30/09: Reception Officially Opens Phase III Expansion
The succesful completion of the HPC Center Phase III expansion was
recognized today at a reception held outside of the new computer room
in Larsen Hall. In attendance was UF President Dr. Machen and
Vice President for Research Dr. Phillips. Each gave a brief statement
recognizing the contributions and importance of the HPC Center to research
at the University of Florda.
03/17/09: Budget Principles Approved
The UF HPC Center budget principles have been approved by the UF HPC
Committee. The budget principles
are in PDF format.
02/25/09: Updated Policies
The UF HPC Center has updated their policies concerning usage. Please
review them on the policies
page.
01/29/09: Phase III, Long Jobs, Scheduling
As many of you may have noticed, the 896-core Phase III cluster has been
running jobs for several weeks now. It is, however, still in a
pre-production testing phase while we shake out some additional minor
hardware problems (mostly questionable DIMMs).
With the addition of the Phase III cluster we are also fine tuning
scheduling policies to insure that we meet the quality-of-service
commitments to our investors while still distributing resources as
equitably as possible. One of the biggest impediments to achieving
these goals is long jobs. Although we recognize the need to run the
occasional long job, it must be recognized that large numbers of long
jobs constrain resources in such a way that some jobs may suffer
inordinately long delays before being run. We would like to reduce
the likelihood of this happening as much as possible.
For some time now long jobs, previously defined as jobs of 7 days or
longer, have fallen into the "longq" which has a limited number of
processors (256) devoted to it. In this way, we can limit the number of
cluster-wide resources devoted to such jobs without prohibiting them
entirely (as many sites do). We have modified the scheduler so that
non-investor jobs will fall into the longq if they are 3 days (72 hours)
or longer. Jobs of 7 days or longer will still fall into the longq
regardless of who submits them. Based on our statistics this change
will have almost no impact on jobs from faculty investors. It will
however impact some non-investors who have a tendency to submit longish
jobs. Such users should reduce their walltime requests where possible.
Note that if you are in the habit of just asking for the maximum amount
of wall time without giving it any thought, you are severely hindering
your ability to run jobs on the cluster.
01/29/09: Job Priority
Insight on how job scheduling priority is attained can be found
on
the wiki
01/28/09: Submission Queue Work
The UF HPC Staff are currently working on the submission queue system. This may result in some delays from
time to time while we are testing new features.
01/21/09: nVidia Tesla unit available for testing
The UF HPC Center now has an nVidia Tesla compute unit available for testing
code written in the CUDA
framework. It can be used to streamline the programming of parallel systems
in specific circumstances. For those users who are developing new software
or are willing to go back and modify their old software in order to take
advantage of this fast computational hardware, let us know if you would
like to work with this hardware and we can get you setup for working on it.
