Case Study

SC11 Student Cluster Competition

Student Competition SC11 logo

OCZ Enterprise SSDs Used in Winning Student Cluster Competition Design at Supercomputing 2011

National Tsing Hua University Team Leveraged OCZ Deneva 2 SSDs to Speed Storage Deployment and Increase Performance of the Application Benchmark Datasets Tested

Background

One of the exciting events held annually at the Supercomputing Conference (SC) in Seattle is the Student Cluster Competition (SCC). For the past five years, this competition has provided a forum for small teams of students around the globe to build high-performance hardware computing cluster systems directly on the SC exhibit floor, where a winning team is crowned champion by demonstrating the best sustained performance across a series of high-performance computing (HPC) applications. For this year’s competition, each student team was tasked to run a set of scientific applications on their cluster computing designs which covered molecular design simulation analysis, computational cosmological expansion, computational biological analysis, and ocean circulation analysis, and the results of these application datasets were then submitted for judging.

Eight university teams, each comprised of six undergraduate students and a coach, were selected to compete in this year’s competition and build the most powerful cluster computer systems they could within a power ceiling limit of 26 amps. The teams had a few months prior to the competition to determine a system configuration, become proficient on how to build it and tune it in real-time, and had to learn the four scientific applications for which they would be judged. Once the competition began, each student team was on their own to build their pre-configured system in full view of SC show attendees, without help from their coach, industry experts, or vendor sponsors.

As part of building their computer cluster systems, hardware and software choices were completely determined by each team and whatever equipment they could obtain from vendor sponsors in advance of the competition. Each team was judged on how their design addressed the computational challenges of the application areas. In addition, each team was interviewed to assess how well the students understood their cluster design, the associated application workloads, and what they learned throughout the process. The team that achieved the highest overall LINPACK score (based on HPCC benchmarks) would become the SCC champion.

The Scientific Applications

The heart of the competition typically began when the student teams received the application datasets that needed to be tested for the four scientific applications, with approximately two days working around the clock to build their systems, number-crunch the data, and complete the testing. Since each application had different operational characteristics and resource demands, it was up to each team to address any system or application problems, as would be common in a data center environment, so the choices made by each team were crucial to success, or failure.

For SC11, the contest organizers chose the following HPC scientific applications to be tested:

MD Simulation Analysis: PFA

Pretty Fast Analysis (PFA) is a software suite for analyzing large-scale molecular dynamics (MD) simulation trajectory data on massive parallel supercomputers, graphics processing unit (GPU) based clusters and traditional GPU-accelerated desktops.

Computational Cosmology: GADGET

GADGET is widely available (and free) code used for performing cosmological N-body/smoothed particle hydrodynamic (SPH) simulations on massive parallel computers and is used for studies of isolated systems or for simulations such as cosmological expansion of space.

Computational Biology: MrBayes

MrBayes is a software program used to help determine the evolutionary process of trees through Bayesian inference, and in particular to approximate posterior probabilities as well as to analyze nucleotides, amino acids, and morphological data models.

Ocean Circulation: POP

POP is widely available (and free) code used as an ocean circulation model that solves three-dimensional primitive equations for fluid motions, and can predict the evolution of ocean properties such as sea-surface height variability.

OCZ's Involvement

Deneva 2 3.5 Inch SSD

OCZ Storage Solutions, a leading provider of high-performance solid-state drives (SSDs) for computing devices and systems, partnered with Tier 1 system OEM, Acer, to support the National Tsing Hua University team of Taiwan – and the returning champion from last year’s competition. Acer supplied the team with their Gemini 72-core Intel Xeon-based cluster system while OCZ provided six Deneva 2 enterprise-class SATA III SSDs for high-speed storage, each supporting 3.5-inch formats and 128GB of storage capacity. NVIDIA also sponsored this team and contributed six Tesla GPU cards.

The Deneva 2 SSDs used by the National Tsing Hua University team each supported up to 80,000 random write IOPS and 275 MB/s of write bandwidth making them well suited for write-intensive HPC environments in servers, cloud computing, and data centers. With capabilities that include power loss data protection, best-in-class endurance (e.g., minimal write amplification, intelligent block management and wear-leveling), and advanced encryption and error correction coding (ECC), these drives are optimized to specifically address the unique needs of enterprise clients.

OCZ Sponsored the Winning Team

The student team from National Tsing Hua University once again claimed first place in the Student Cluster Competition at the Supercomputing Conference, having produced outstanding results on their x86-based hardware and their clustered computing tests and benchmarks. The team’s final configuration using Acer’s high-density Gemini cluster system with NVIDIA Tesla C2070 GPUs optimized the number-crunching process (depending on the scientific application tested) and achieved over 1.8 TFLOPS across six systems with a power efficiency under 5.7kW. OCZ’s enterprise class Deneva 2 SSDs were used to speed the storage deployment and enabled fast input/output operations per second (IOPS) performance of the scientific application benchmarks required for review and analysis by the student team.

DOWNLOAD PRINTABLE VERSION (PDF)