Straight-forward, familiar software environment

Designed for the Programmer

  • Simple API (in C) for initiating and controlling data transfers low-level, but small and efficient
  • Flat network model, all nodes are equidistant
  • Linear scaling usually comes for “free” develop code on 4 nodes; run on 4, 8, 16, 32, 64 nodes
  • Superior performance on GUPs, FFT and Graph500 BFS

Programming Methods 

  • Custom Code with Low Level AP
  • DV (or customer) developed Libraries
    • 3D FFT now
    • More to follow later
  • Higher level API
    • Send/Receive available soon
    • Put/Get under development
  • HYBRID code
    • Run MPI code on DV20x without DV VIC/Switch
    • Optimize time-critical sections to run on DV VIC/Switch

High performance with a simple and natural programming model

The software to control the Data Vortex® network switch is easier to program and more flexible than software for other types of switches. Because the Data Vortex network switch can be integrated into an industry-standard commodity cluster, it provides a familiar software environment using the industry-standard C programming language.  A rich application programmer’s interface (API) is provided to enable transferring data to and from the network switch. It is conceptually similar to message passing, but designed for SRAM and remote memory which provides advantages like global addressing and better performance.

Some of our proven software applications include FFTs, Sparse Matrix Linear Algebra, and scatter-gather operations. These algorithms don’t scale well on current supercomputers. Data Vortex Systems are designed to solve Hard Problems by performing and scaling on such algorithms. And unlike many other advanced computing systems, the Data Vortex programming model is the same no matter the size of the system.  

Thunder_Horse_SemisubHard Problem: Oil & Gas Exploration

FFT algorithms are used to analyze enormous seismic data sets, reduce processing time, and increase efficiency and accuracy of exploration.

Hurricane_Fran_sept_1996Hard Problem: Climate Modeling

Sparse Matrix Vector Multiply algorithms are used to predict climate effects on weather, farming, and water management.

 

Current Data Vortex Research

Any problem with a lot of communications or even just a lot of small data transfers like high frequency trading, life sciences, big data graph analytics, and much more. Examples of these problems currently being run on Data Vortex Systems can be found on our RESEARCH page.