Millions on the core of the parallel to make you become "faster"

How to use a 4-core processor for a desktop or notebook, and for those of us who don't quite understand it, 4 and processors are simply waste. Of course, this is not our problem, in the writing parallel execution of the code is very difficult, so most of the application (except for some computational intensive video games) can only run on a single core. However, recent a group of researchers from Stanford have released parallel processing at the million core.

As parallel processing is well known to more and more people, running applications on multiple nodes can be faster than ever before, so your workload can be allocated to multiple nodes for parallel processing, and the more core the faster. That's why Hadoop was born, and it's good for big-volume data processing. The MapReduce framework assigns jobs to multiple nodes for processing and then summarizes the resulting output.

But even Hadoop can only run on tens of thousands of nodes, because it focuses on nodes-not very much on the use of a core processor. IBM-produced Sequoia supercomputer (housed at the Livermore National Laboratory), the Stanford team is made up of 98,304 processors (or nodes), each processor encapsulates 16 compute cores, totaling 1572864 cores; Researchers have been able to use most of the core, They say they create records on many categories.

Without the record, this is indeed a complex and incredible task; This simulation enables more than millions of tasks to be allocated on tens of thousands of nodes (16 cores per node). Because if a collision occurs in these processes, it can cause the system to slow down or destroy the entire simulation.

Even in the hundreds of thousands of-core world of the universe, where there may be special-purpose processors with GPU, there is a lack of solutions to make the most of these cores. The world of High-performance computing is moving toward millions of megabytes, but problems that are bigger than energy consumption may arise-applications need more computing power and algorithms that are adaptable to this size.

"Edit Recommendation"

IBM2012 Annual earnings Interpretation: Cloud computing business rose 80%joyent and Skytap launched the Cloud Hadoop service gartner:2015 Year 65% advanced analysis products will be embedded Hadoopapache Hadoop cluster security Architecture "The Executive editor: Xiao Yun TEL: (010) 68476606 "