Visit additional Tabor Communication Publications
January 22, 2010
Like just about everyone else, I head into the SC conference season hungry for cool new hardware news. Throw in a quantum computing demonstration, or string together a couple thousand Atom processors into a computing prototype, and I head for the keyboard like a suburban American dad heads for the turkey on Thanksgiving day.
There are other kinds of announcements that happen during SC about new services, business models, software, machine installations, and so on. So many announcements, in fact, that it's not possible to do more than perfunctory service to most of them. This article is about one of the important developments I didn't cover during the November rush: the announcement of Mathematica 7.
On the day the exhibit floor opened at SC08, Wolfram Research announced the latest version of its flagship product, Mathematica 7. The tool has long been a part of the "thinking workflow" for engineers, scientists, and a host of other technical professionals. Mathematica allows uers to manipulate mathematical statements, and visualize the results, using mathematical notation directly rather than a special grammar or intermediate form. What is important about the Mathematica 7 announcement is that it marked the introduction of features that turn Mathematica into a gateway through which users can take advantage of the power of high performance and large-scale computing without changing their workflow and in many cases without being aware that they are working in parallel.
Mathematica is a symbolic manipulation environment that allows users to manipulate mathematical and logical expressions directly in the mathematical notation you'd expect to find in a textbook or use with pen and paper. Alongside this capability for symbolic manipulation lie sophicated numerical methods that allow the system to reduce expressions to numeric answers when appropriate. Reports from users of several of the most popular systems indicate that Mathematica, always strong with respect to symbolic manipulation, has closed the numerical gap it once had with respect to other packages.
I came from a MATLAB-centric environment, and so I was interested in knowing how the tools are different from one another. In contrast with Mathematica's focus on expression in symbolic language, MATLAB offers users a language that is similar to C and HPF, a distinction that reflects the fundamental difference in the kind of thinking the two tools are designed to support. There is a widely-held generality that if your interest is in numerical methods and matrix manipulation, MATLAB is the right choice. On the other hand, if you are more generally interested in mathematics, or need to express theoretical problems directly in the language of mathematics, Mathematica is the right choice. Both packages are quite sophisticated, and I find that this generalization isn't always helpful.
One user I spoke with described the choice between MATLAB and Mathematica as one that comes down to how your brain is wired: if you naturally start with equations on paper and then work your way out to data and programs, then Mathematica is likely to be a good fit for you. But if you like to start with data first and then try to find a model that fits what you observe, MATLAB is the way to go.
Either way, these environments and others like them offer users the ability to use a computer to answer questions or solve problems without having to reduce the problem to an expression in C or FORTRAN. And many of the available mathematics environments, MATLAB and Mathematica among them, offer rich support for parallel processing and a wide variety of third party tools.
But why does this matter?
Researchers using computer applications to facilitate their research can expect to face two discontinuities as the computational requirements of their applications grow and they require more computer resources. The first discontinuity occurs when an application moves from operating with a single thread of execution to multiple simultaneous threads. Today this transition can happen while the user stays in the familiar operating environment of his desktop, but he may also transition to a small shared resource. This shared resource could look a lot like his familiar desktop environment, or it might be a totally new environment that disrupts his existing workflow and toolsets. The application usually has to be rearchitected to take advantage of the performance potential offered by adding more execution threads, and this can require substantial new skills or even the addition of new team members to parallelize the application.
The next transition is to resources that permit simultaneous execution of a very large number of tasks: supercomputers. This often involves rearchitecting the application again to take advantage of orders of magnitude more parallelism. Unfortunately, the transition also nearly always involves a complete disruption of previous workflows, new tools, and new ways for the user to interact with the resource during development. Literally everything may be new, from the command line interface (no Gnome or Windows GUI on most supercomputers) to working with a batch queuing system. This is a problem for users who must adapt to an unfamiliar and often unfriendly environment at the same time that they are dealing with the fundamental issues of the science, and computer science, in their application that frequently arise at supercomputing scale.
The features included in last November's upgrade of Mathematica are designed to minimize both kinds of workflow disruptions. Support for parallelism is now included in all versions of the product, reflecting the growing ubiquity of multicore processors. The approach, which the company calls "immediate interactive parallel computing," builds multicore awareness into the base application so that when an opportunity arises to run on multiple cores, the application will automatically do that. No change to the expression of the problem is necessary, and no checkboxes have to be ticked by the user. Mathematica just takes advantage of multiple cores on behalf of the user when it makes sense to do so.
Mathematica also supports semantics for explicitly-declared parallelism. There are parallel analogues to serial functions such as Table and Map that allow users to specify that certain portions of a computation be performed in parallel. There are also commands that allow users to define a function and then distribute that definition to multiple processors for parallel evaluation. In both cases Mathematica handles the generation, scheduling and rendezvous of threads and gathering of results automatically so that the user does not have to specify that level of detail.
This screencast on the Mathematica site offers an excellent introduction to explicit parallelism, and provides a window into how Mathematica works. You'll find many other screencasts and live examples on the site as well if you want a more detailed look at how Mathematica works.
But what about growing beyond the cores in a socket? Wolfram added gridMathematica in this release to provide parallel semantics to allow users to access everything from multiple sockets in a single box to HPC clusters. gridMathematica is licensed in two versions, Local and Server. The local version adds another four cores to the four licensed in the standard version of Mathematica, and will let users take advantage of multiple sockets on a single machine. The server license comes in bundles of 16 cores, and can support larger scale processing. gridMathematica can integrate with common batch workload schedulers, but it doesn't have to.
Wolfram has created what it calls the Lightweight Grid System to manage cluster resources as you might need if cycles are being harvested from resources that are usually used for something else. Interestingly gridMathematica supports clusters with heterogeneous nodes: provided Mathematica supports a particular platform or OS, gridMathematica can harness it into a single cluster. This means that clusters can be built from any mix of 32- and 64-bit platforms running Windows, Mac OS, Linux, and Solaris.
Wolfram has expanded this concept even further with a planned feature called HPC Cloud Service for Mathematica. This feature will allow users who don't already have a cluster to run their Mathematica application on a commercial cloud resource. Users will take advantage of the cloud service by clicking a few buttons in the cloud GUI with their existing Mathematica notebook to ask it to run the notebook in the cloud.
The concept is that Mathematica automatically analyzes your notebook, creates the bundles of work and runs them in parallel, gathers the parallel results, and sends them back to the desktop where the user initiated the request. Originally announced in November, Wolfram's HPC Cloud offering has been delayed by the company's Wolfram|Alpha computational knowledge engine. No word on when release is expected, although the company does emphasize that work is continuing on the HPC Cloud Service offering.
So where does Mathematica fit in your environment? If you, or your center's users, are primarily using mature Fortran or C applications that have been under development for many years, it may not fit at all. All of these environments take time to become proficient, and Mathematica is no exception. On the other hand, if you already have users of mathematical environments of one flavor or another, or you are accommodating users who are developing new applications and who are domain specialists rather than "HPC people," Mathematica may be a good fit. It has the potential to increase the effectiveness of your users and provide them with higher levels of performance in a way that is much easier to manage than moving directly from an idea (or a working serial prototype) to an explicitly-parallelized traditional application.
May 23, 2013 |
The study of climate change is one of those scientific problems where it is almost essential to model the entire Earth to attain accurate results and make worthwhile predictions. In an attempt to make climate science more accessible to smaller research facilities, NASA introduced what they call ‘Climate in a Box,’ a system they note acts as a desktop supercomputer.
May 22, 2013 |
At some point in the not-too-distant future, building powerful, miniature computing systems will be considered a hobby for high schoolers, just as robotics or even Lego-building are today. That could be made possible through recent advancements made with the Raspberry Pi computers.
May 16, 2013 |
When it comes to cloud, long distances mean unacceptably high latencies. Researchers from the University of Bonn in Germany examined those latency issues of doing CFD modeling in the cloud by utilizing a common CFD and its utilization in HPC instance types including both CPU and GPU cores of Amazon EC2.
May 15, 2013 |
Supercomputers at the Department of Energy’s National Energy Research Scientific Computing Center (NERSC) have worked on important computational problems such as collapse of the atomic state, the optimization of chemical catalysts, and now modeling popping bubbles.
05/10/2013 | Cleversafe, Cray, DDN, NetApp, & Panasas | From Wall Street to Hollywood, drug discovery to homeland security, companies and organizations of all sizes and stripes are coming face to face with the challenges – and opportunities – afforded by Big Data. Before anyone can utilize these extraordinary data repositories, however, they must first harness and manage their data stores, and do so utilizing technologies that underscore affordability, security, and scalability.
04/15/2013 | Bull | “50% of HPC users say their largest jobs scale to 120 cores or less.” How about yours? Are your codes ready to take advantage of today’s and tomorrow’s ultra-parallel HPC systems? Download this White Paper by Analysts Intersect360 Research to see what Bull and Intel’s Center for Excellence in Parallel Programming can do for your codes.
In this demonstration of SGI DMF ZeroWatt disk solution, Dr. Eng Lim Goh, SGI CTO, discusses a function of SGI DMF software to reduce costs and power consumption in an exascale (Big Data) storage datacenter.
The Cray CS300-AC cluster supercomputer offers energy efficient, air-cooled design based on modular, industry-standard platforms featuring the latest processor and network technologies and a wide range of datacenter cooling requirements.