Technology:Best Work-Sandia-Computational Science Research

Computational Science Research Highlights

Algorithms and Mathematics

• New Partitioning Algorithms Reduce Communication Overload in Parallel Computing - "As part of research supported by the CSCAPES (Combinatorial Scientific Computing and Petascale Simulations) SciDAC institute, we have developed new sparse matrix partitioning algorithms that use a 2D partitioning… . We believe our data partitioning will be useful in a variety of algorithms, not just matrix-vector multiplication."
  
  
• Algebraic Multi-grid Algorithms Scale to Thousands of Processors - " Our research focuses on new algorithms that differ significantly from traditional AMG in processing coarse models much more efficiently. This has allowed for unprecedented performance on notoriously difficult applications."
  
  
• New Algorithms Improve Performance on Next Generation Multi-core Computers - "Our efforts are focused on the development of algorithms that take advantage of lower-precision data (single-precision vs. double-precision floating point data), higher computation-to-bandwidth ratios, and hybrid distributed-shared memory algorithms."
  
  
• The New Moore’s Law and Sandia’s Role in the Future of Computing "…the physics of computation is an unexpected new force demanding a rewrite of old code… . This project has identified the source of this change in physics and is predicting Sandia’s future role systematically rather than on an ad hoc basis..."
  
  "...Computer scientists have been very successful in finding algorithms that minimize the number of arithmetic operations (ops/FLOPS) needed to solve a particular type of problem, such as matrix multiplication. This was very relevant in the past when logic gates were precious, but is less relevant today now that power consumption is more important..."
  
  "...The future vision is for computer chips that are augmented by new layers of nanodevices that perform data storage and highly power efficient but specialized logic functions..."
  
  
• Comparison Reveals Relative Strengths of Multithreaded and Distributed Memory Architectures -"...Our results show both that distributed memory architectures are viable, even on power-law data, and that massively multithreaded architectures can be competitive doing floating- point computations..."
  
  
• Enhancing Next Generation Interconnection Networks for High-Performance Computing
  "…it is critical that each new generation of machine continues to improve message passage performance in order to enable high levels of scalability. Our research addresses how this can be done in an environment where realized MPI performance is increasingly difficult to achieve."
  
  
• Catamount Lightweight Operating System Maximizes Multi-Core Compute Resources While Minimizing Power
  "Our work on the Catamount lightweight kernel increases the scalability performance, effectiveness, and efficiency of the parallel computing platforms used for delivering advanced modeling and simulation solutions for science-based engineering."
  
  
• Allocation Algorithms Maximize Throughput in Parallel Supercomputers
  "...The CPA (at less than one percent of the cost of a parallel computer) is an example of how a relatively small investment in computer algorithms can dramatically leverage the return on a large investment in computer hardware..."
  
  
• Kitten Lightweight Kernel Makes Multi-Core Processors Purr
  " Our next-generation open-source LWK, called Kitten, is designed specifically for current and future multi-core processors, which will likely form the basis for upcoming exascale computing platforms."

Technology:Best Work-Sandia National Labs Accomplishments

Computing and information sciences

• Red Storm was successfully upgraded to include 38,400 processing elements through the introduction of new quad-core AMD Opteron processors.
• The peak processing power of Red Storm rose to 284 trillion operations per second
• It also is in the top 10 of the TOP500 list
• developed a new hash algorithm and submitted it to the National Institute of Standards and Technology for consideration as a new national standard.
• Hash functions are extremely important for authenticating data. We use one every time we make an Internet purchase, for example. Recently the current standard family of hash functions used for Internet authentication was partially compromised by Chinese mathematicians, prompting the call for a competition to find a better solution.
• Sandia convened a meeting in Washington of 85 leaders from industry, government, and academia to confirm the need for computation-based engineering (CBE) as a tool for US competitiveness.

Technology:Sandia supercomputer boots a million virtual machines

• Sandia supercomputer boots a million virtual machines. Computer scientists at the Energy Department's Sandia National Laboratories have simultaneously booted one million Linux kernels, all of which ran as virtual machines on the labs' Thunderbird supercomputer.


• Thunderbird is a 4,480-node Dell-based computer cluster. Each node ran 250 Linux kernels. The host OS on each node is a stripped-down version of Linux kernel, compiled by the researchers themselves. It contains only the kernel core and and a start-up script that boots up the virtual machines. "...The root file system lives out in Random Access Memory..."
• For the virtualization, the system uses a hypervisor built into the Linux kernel, called Lguest, which was developed by the research arm of IBM. ...it is "very fast and very lightweight,"...the start-up for each virtual machine is a fraction of second..."

• The management software is OneSis, which was originally developed by Sandia. "...OneSis is pretty key to making this thing work at all. It is good at managing thousands of thousands of nodes in a very easy way..."