Visit additional Tabor Communication Publications
March 16, 2007
For the past year or so, Intel has been talking up RMS, an application software model for terascale computing. RMS, which stands for Recognition, Mining and Synthesis, is the class of software that Intel believes will represent the killer apps of the not-too-distant future -- 2010 and beyond. These applications will require the capabilities of terascale processors -- teraflop performance processing terabytes of data.
RMS applications are used to manipulate complex models, which can be either objects or events. The three components of RMS describe its function. Recognition involves defining the model; mining has to do with sifting through datasets for instances of that model; and synthesis performs "what-if" types of calculations on the model to yield predictions or solutions.
For example in financial markets, there is a deluge of real-time data being generated for all types of traded options (bonds, stocks, currency). This makes qualitative financial analysis somewhat of a guessing game. RMS might change that. A trader would be able to define a model of an attractive options investment (recognition). Once an investment of this type is found (mining), an algorithm could be applied to provide the trader with the different levels of risk and potential return (synthesis) in a given economic environment. The application doesn't provide the trader with an answer, per se, just guidance on which investments are more likely to yield good returns under different sets of economic conditions (interest rates, currency rates, stock P/E ratios, etc.).
Other examples can be found in a variety areas, such as medicine (cancer treatment), security (terrorist threat identification), communication (real-time language translation), retail (virtual apparel shopping), and transportation (automated vehicle control), to name a few. In fact, it would be hard to find an area of human endeavor that would not be able to take advantage of RMS systems. After all, virtually everything we encounter in the real world can be modeled.
"It's a very powerful paradigm for dissecting problems," says Jerry Bautista, Director of Technology Management for Intel's Microprocessor Research Laboratory and co-director of Intel's Terascale Program. "It has tremendous application in almost anyplace you look. We're trying to use the computer to suggest a course of action. That's a lot more powerful than Google, where you're just trying to find something."
The fact that RMS application must manipulate whole models, rather than lower level constructs, not only means that more computing power is required, but also more sophisticated software.
"The problem is ordinary computers don't model things," said Pradeep Dubey, Senior Principal Engineer, Manager of Innovative Platform Architecture Microprocessor Technology Lab, Intel. "Aside from supercomputers, today's computers aren't capable of developing mathematical models of complex objects, systems or processes."
An important feature of this model is the interactive nature of the applications. This works on a couple of different levels. The first is that the types of applications being envisioned usually have to deal with real-time data input, and then turn around and generate timely output. For the options trading application above, it means that the investment predictions must be timely; an opportunity may no longer be relevant after a few hours, or perhaps even minutes.
Another aspect to the interactiveness is the ability of the software to learn, through the feedback of the mining and/or synthesis operations. Again using the example of the options trading app, the application could track actual performance of trades it identified and use that information to refine the original model of what defines an attractive investment.
"In this new approach, it doesn't matter how good the model is to begin with," explains Dubey. "The use of real-time feedback loops can make the software much more powerful than static applications."
If this is starting to sound like Artificial Intelligence (AI), it's because it is -- or was. Intel tends to shy away from such jargon, since the AI term has become rather loaded. But the idea of an application suggesting outcomes or solutions is more akin to what humans do. It's been the Holy Grail of computing for 50 years. Intel's Terascale Program can be seen as the company's attempt to encapsulate this classic goal in a new way.
"The Terascale Program is not the case of Intel just flexing its hardware muscle and deciding this is the next thing to do," says Bautista. "We actually started this RMS research several years ago, and we were driven by the applications themselves. We thought that they were compelling and had broad applicability -- not just for high-end users but also for normal people to use in their homes."
Certainly, if people were able to buy systems capable of such things as natural language recognition or image-based data mining, they would be in demand today. For these applications to be useful, nothing else needs to happen -- except for the systems to be invented.
According to Bautista, Intel's RMS research is very well integrated into the company's overall technology strategy. They're not off working in a corner by themselves, he says. The RMS development is driving the direction for a variety of hardware technologies looking three to seven years out.
The nature of model-based computing points to certain hardware capabilities. Some of those capabilities are embodied in Intel's 80-core terascale processor prototype, demonstrated recently at the Integrated Solid State Circuits Conference. The basic idea is to compute in a highly parallel fashion and move data around very quickly between processors, memory and external devices.
Parallelism is a big part of it. RMS workloads have an insatiable appetite for computing power. Forget dual-core. Applications that manipulate models demand manycore architectures, supporting hundreds of threads. An RMS program will need the computational performance of a supercomputer or entire datacenter.
Another requirement is high processor-to-memory bandwidth. Intel is considering several approaches to scale the memory wall problem. The first is to place memory chips in the same package as the processor. Another approach is to stack memory chips underneath the processor -- something the company has suggested it will do in a future prototype of its terascale processor. Lastly, memory can be placed on the chip, something already done on a relatively small scale with cache memory.
Since the RMS apps will access large dynamic data sets, high performance I/O is another key requirement. Intel has produced early demonstrations of silicon-based photonics, a model that could be used to increase I/O communication by a couple orders of magnitude (terabits of data per second). The only question mark for silicon photonics is if the technology can be implemented economically on standard semiconductors.
While the production of general-purpose terascale hardware in one form or another is almost a sure thing within the next five years, the software is another matter. The highly parallel nature of a future teraflop processor means that new classes of software and software technologies will have to be developed if the average programmer is going to be able to write applications for them. Parallel computing represents a paradigm where the hardware pretty quickly outruns the ability of developers to program it. That's why Intel is starting to think about these issues today.
The company is partnering with other organizations to move the effort along. Intel has been awarding research grants to a number of academic institutions to work on the fundamental problems associated with RMS software. In addition they've been seeding advanced software curriculums (i.e., courses in parallel computing) at over 40 colleges and universities, and are looking to add 100 more institutions this year. According to Bautista, the RMS work at Intel has progressed more rapidly than they anticipated, so they've become more aggressive in getting academia to catch up. Now that hardware prototypes are available, he expects the work will begin to accelerate.
In addition, Intel's Software Solutions Group is actively collaborating with Microsoft and others in the ISV community to help them prepare for the new application model. According to Bautista, ISVs are quite enthusiastic about terascale applications and the model-based paradigm.
"To be honest," says Bautista, "the discussion usually end up with them saying: 'When can we get this thing? If I had it, I'd use it today.'"
Jun 19, 2013 |
Supercomputer architectures have evolved considerably over the last 20 years, particularly in the number of processors that are linked together. One aspect of HPC architecture that hasn't changed is the MPI programming model.
Jun 18, 2013 |
The world's largest supercomputers, like Tianhe-2, are great at traditional, compute-intensive HPC workloads, such as simulating atomic decay or modeling tornados. But data-intensive applications--such as mining big data sets for connections--is a different sort of workload, and runs best on a different sort of computer.
Jun 18, 2013 |
Researchers are finding innovative uses for Gordon, the 285 teraflop supercomputer housed at the San Diego Supercomputer Center (SDSC) that has a unique Flash-based storage system. Since going online, researchers have put the incredibly fast I/O to use on a wide variety of workloads, ranging from chemistry to political science.
Jun 17, 2013 |
The advent of low-power mobile processors and cloud delivery models is changing the economics of computing. But just as an economy car is good at different things than a full size truck, an HPC workload still has certain computing demands that neither the fastest smartphone nor the most elastic cloud cluster can fulfill.
Jun 14, 2013 |
For all the progress we've made in IT over the last 50 years, there's one area of life that has steadfastly eluded the grasp of computers: understanding human language. Now, researchers at the Texas Advanced Computing Center (TACC) are utilizing a Hadoop cluster on its Longhorn supercomputer to move the state of the art of language processing a little bit further.
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.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?
Join our webinar to learn how IT managers can migrate to a more resilient, flexible and scalable solution that grows with the data center. Mellanox VMS is future-proof, efficient and brings significant CAPEX and OPEX savings. The VMS is available today.