Campus “Rocks” Computational Cluster
University of California, Santa Cruz
Submitted by:
Stephen Hauskins
Director of Computing for
The Division of Physical and Biological Sciences
University of California, Santa Cruz
Team Lead: Terry Figel
Team Members: Tomas Witten-Hannah, Tim Gustafson
Student Assistants: Melisa Bernetsky, Colin Fahy
Summary:
The Santa Cruz campus has for many years lacked a coherent and available way for doing cluster computations locally. Our need was for a computation cluster that could support the campus as a whole with multiple software packages and sufficient computational capabilities. Terry Figel, the ITS assigned computing director for the School of Engineering, obtained equipment via a donation that resulted in the creation of a very capable cluster that is available to the entire campus for virtually any research computational use.
Project Description
It is important to keep in mind that this was not a formal project in the sense of having a documented path that included milestones and identified project leads. It was an informal process conducted by Terry Figel and his support staff.
Their technical background along with the equipment donation created an excellent combination to have a cluster built and implemented as an informal internal project. The equipment used in the initial creation of the cluster was a donation from Craigslist, the commonly used free website that is known worldwide. The equipment that was donated was a perfect fit for what was to become a fairly high powered computational cluster.
The campusrocks cluster primary use to date has been research in Physics, Chemistry, Computer Science and Engineering. It’s utilization has fluctuated between 50 and 100 percent
This website gives an overview of the Campus Rocks Cluster:
Initial Technology used for cluster
●Rocks cluster software (
●SunGrid queueing job (
●60 nodes
●16GB of RAM per node
●50% of the nodes are 2 cores
●50% of the nodes are 4 cores
●Total of 180 processors
●Scientific computational software as supplied by clients
●Basic compilers and Linux tools
●2 file servers mirrored with 3 TB each running opensolaris and ZFS, using rsnapshot for backing one up to the other.
●Utilizes an LDAP server set up in School of Engineering that is serving VPN, and wireless access for the campus.
Additional Technology
●File Servers upgraded to 11TB each July 2010
●Two 8 core nodes added January 2011
●Two 48 core nodes added April 2011 (bringing the cluster up to 300 cores)
Time Frame
The original machines were donated in the Summer of 2009. Original discussion slowed the roll out based on what authentication, and the trepidation on rolling out a service with no financial backing. The cluster was put together August of 2009, but did not get campus users until the Winter of 2010. Until the recent addition of the 2x48 core servers, it was all done with donated equipment.
Objective customer satisfaction data
This may not sound very impressive to many people, but for my client base located in the Division of Physical and Biological Sciences this was a “major” step forward in being able to offer this type of service to many of our clients in need of a large scale computational service. Prior to the introduction of the cluster I attempted to supply what I could via small dual and quad core servers. This was a step up for many doing their computational work on desktops, but it was extremely limited in terms of having multiple simultaneous users and cpu’s, as well as substantial disk space. In addition, we have seen computational speeds increasing a 100 fold. - Stephen Hauskins
------
My name is Chad Renzelman and I am a Doctoral Candidate at the University of California Santa Cruz, working on my PhD Thesis in Organic Chemistry. Not too long ago I began collaboration with Prof. Jaime Rodriguez from the Universidade de Caruna, in Spain. We began to envision a new method of determining the 3D conformations of cyclic peptides in a low dielectric medium such as chloroform to represent the molecule as it exists while passing through a membrane. We planned on using Nuclear Magnetic Resonance in particular to measure the experimental heteronuclear coupling constants and then compare them to a database of theoretical heteronuclear coupling constants calculated from large set of modeled amino acid rotamers. So I began to work on the calculations using my laptop at first. I soon found that it took 3 days to perform one calculation and I had to perform many thousands. I had a peptide with 6 amino acids, each one has 1296 possible rotamers each one would give a different heteronuclear coupling constant value, so in order to produce my database I needed to perform 7776 calculations, each one involved two different calculation steps that are performed consecutively. So I turned to the IT department for help! I am an organic chemist I don’t have any experience in programming or computer science apart from hacking old video games in high school, but the access to the campusrocks cluster has dramatically affected my research for the better! I have been able to run all of my jobs without a problem, this has allowed to me to develop a novel method of NMR 3D structure determination. On the side I have used the cluster to solve the absolute stereochemistry of a natural product from the Crew’s Lab that was previously unknown and there WAS a controversy over. So without Campusrocks my thesis wouldn’t be half as important to the scientific community as it is now! I would like to thank Craigslist for the wonderful donation to the University! As a modern scientist I find that more and more research can be dramatically enhance through the use of valuable computing resources like Campusrocks. I hope that my research helps illustrate the need for such services, and I hope i can help justify the need to continue support of this invaluable service in the future.
Chad M. Renzelman
Graduate Student Researcher
------
First, I should say that both Auditya Sharma and myself are in the same
group, so the jobs submitted by him and by me serve the same project.
The campusrocks cluster is very useful to our work. The jobs we run on the
cluster are exclusively quantum Monte Carlo simulations of many body spin systems. In our project we investigate mostly the ground state of such systems in the hope that it would eventually give us some information about the advantages that future quantum computers would have over classical ones.
Itay Hen - Physics
Paraphrasing one client
Previously I was using a small quad core server supplied by the Division, it did the job but was not totally adequate for the type of calculations I was performing. Now with the campusrocks cluster I am seeing large increased in computational speeds and I am able to running larger more complex simulations doing Quantum Mechanical simulations.
William James McDonald
Graduate Student