1. What HPC is
As of July 3rd, 2018, the HPC facility consists of a 271 node, 10272 core Condo Cluster available to all researchers at UCI (no School or Dept affiliation required). Condo means that most of the compute nodes are individually owned by contributing Schools, Depts, or individual PIs, and are available to them preferentially, but when not so used, are available to the rest of campus users, using the Free Queue system. This allows all users to effectively get more computational power than they paid for, since usage among owners is quite spiky and the Free Queue allows us to harvest otherwise unused cycles.
While the HPC facility is not HIPAA-compliant, we have been told that appropriately anonymized human data are being reduced and analyzed through various pipelines on HPC.
In brief, HPC offers very large computing resources to all UCI researchers to allow them to store and analyze data of a size that would not otherwise be possible locally, and personal assistance in resolving specific computational problems. Should UCI researchers require even larger compute power, Harry Mangalam is an XSEDE Campus Champion, who can help to move your problems to the larger XSEDE resources.
The HPC cluster is now the largest on campus with about 9900 64bit CPU cores, ~55TB aggregate RAM, 30 Nvidia GPUs, with Quad Data Rate (40Gbs) Infiniband interconnects. There is ~2PB of storage with >1PB in 4 BeeGFS distributed filesystems, and the rest available from ~34 public and private NFS automounts. We have ~2000 users, of whom ~100 are logged in at any one time on 2 login nodes and over 250 use the cluster in an average month. The cluster generally has 65%-90% usage with several thousand jobs running and queued at any one time (>6000 as I write this; 20,000 is not unusual).
The HPC staff can usually add requested Open Source Software within a day; our module system has >1000 external applications (various versions), libraries, and compilers/interpreters. These include image processing (freesurfer/fsl), statistics (R, SAS, STATA) numerical processing (MATLAB, Mathematica, Pylab/Numpy/SciPy, Scilab, GSL, Boost, etc), molecular dynamics (Gromacs, NAMD, AMBER, VMD), as well as a number of Engineering applications (OpenSees, Openfoam).
However, since our largest user group is biologists, we have an exceptionally large group of bioinformatics tools including the bam/sam/bedtools, tophat, bowtie, bwa, wgs, trimmomatic, repeatmasker, fastx toolkit, emboss, tacg, Bioconductor, BLAST, Trinity, GATK, Annovar, velvet, vcftools, cufflinks, smrtanalysis, the SOAP tools, picard, the Bio(Perl|Python|Java) toolkits, etc.
Our compilers support parallel approaches including OpenMP, OpenMPI, Gnu Parallel, R’s SNOW, and CUDA. We support debugging and profiling with TotalView, PerfSuite, HPCToolkit, Oprofile, and PAPI.
Our module library consists of this list. Note the date on top for the most recent update.
We run classes and tutorials in using Linux and the HPC cluster as well as Bioinformatics (in conjunction with the Genomic High Throughput Facility) and also offer classes in BigData processing (with Padhraic Smyth of Information & Computer Sciences). You are welcome to view and use the class tutorial.
HPC is supported by ~5 FTEs, including 1 PhD (in Molecular Biology & Bioinformatics), and these support staff regularly assist with debugging workflows and performance problems in all domains (700 domain-specific emails per month). We very much welcome new computational and research challenges and our interaction with scientists. Please feel free to mail all of us at <firstname.lastname@example.org>, altho if you want help on a particular problem, please phrase it so we can help you.
Harry Mangalam, PhD, for the Research Computing Support Staff
Joseph Farran, Senior Sysadmin, 30+ years IT experience
Imam Toufique, Sysadmin, Virtualization & Network Gymnast
Francisco Lopez, Sysadmin, Security, Cloud
'James Walker, Sysadmin
Garr Updegraff, PhD,
Edward Xia (student)