HPC for all in MathWorks double whammy
The MathWorks is making supercomputing more accessible to users by running its parallel computing products on the EGEE grid and improving MATLAB’s parallel applications.
The MathWorks recently announced its Parallel Computing Toolbox will let MATLAB users create parallel apps and distribute them to other users as standalone executables or software components for use on computer clusters. Silvina Grad-Freilich, manager of parallel computing marketing at The MathWorks, said: ‘The code should not have to change between the desk top and using the external resources of, for example, a HPC cluster.’
In addition, the Parallel Computing Toolbox has also incorporated a new language construct, called ‘spmd’ that simplifies the development of data-intensive parallel applications.
An example of this portability of code is the ‘parfor’ loop command within MATLAB, which contains the same code as a ‘for’ loop, but instead tells the code to run over multiple machines. Grad-Freilich added: ‘The user does not need to learn a parallel language when they want to run code over multiple machines.’
Such seamless supercomputing is a major pull for the team at The MathWorks, as Jos Martin, principal engineer at The MathWorks, said: ‘The user could run an algorithm and not realise that it’s running in parallel.’
Furthermore, MATLAB users can now convert parallel MATLAB applications into executables or shared libraries and provide then to their own end users royalty-free. The main benefit of this is that the non-MATLAB end users could be using MATLAB tools without realising it.
So could seamless supercomputing be on the horizon? Loren Dean, director of engineering of MATLAB products, said: ‘When you are printing out a word document, you expect it to look the same no matter what the printer is. And in the same way, you expect code running on a cluster to react the same, no matter what hardware is being used.’
Furthermore, The MathWorks’ parallel computing products – Parallel Computing Toobox and MATLAB Distributed Computing Server – are now running on the EGEE grid (Enabling Grids for E-sciencE). These tools are fully integrated with gLite, the middleware used by EGEE. Prior to this integration, users could not work with MATLAB in a grid environment but now EGEE users can access the power of the grid directly from their desktops.
Martin added: ‘We do not want them [scientists] to be HPC users, we want them to do their science. One of the biggest problems in science is how to use a desktop computer and then how to scale up to a cluster and we are trying to eliminate that problem.’