• 13:22
  • Friday ,19 May 2017

The Machine: HP Enterprise's 'memory-driven' computer for the age of big data




Friday ,19 May 2017

The Machine: HP Enterprise's 'memory-driven' computer for the age of big data

They say it's a computer for the big data era. Hewlett Packard Enterprise has unboxed The Machine, boasting 160 terabytes of memory. But wait! It's just a prototype.

has been developing this computer specifically to capitalize on the endless data every one of us generates. It has the potential to out-pace even the fastest supercomputer on the planet, China's Sunway TaihuLight.

"The secrets to the next great scientific breakthrough […] hide in plain sight behind the mountains of data we create every day," said HPE head Meg Whitman in a statement. "We can't rely on the technologies of the past. We need a computer built for the big data era."
HPE's Machine is based on a new concept called Memory-Driven Computing. That puts memory - and not the processor - at the center of the computing architecture. HPE says it will reduce the time it takes to process complex calculations "from days to hours, hours to minutes," and will ultimately deliver real-time intelligence.  
With 160 terabytes of memory, HPE says The Machine could hold and work with the data in 800 million books. It's called a "single-memory" computer because every processor in the system accesses a giant shared pool of memory.
In standard systems today, each processor is assigned a relatively small amount of memory on which it's allowed to "feed." If it needs more memory, it has to request access from another processor. That wouldn't happen in a single-memory computer, as all processors have equal access and can feed as they need. Allowing all the processors to feed together could cut inefficiencies.
Other common computing names, like IBM and Intel, are developing similar memory-centric technologies or are seeking to address the inefficiencies in computing and memory.
 HPE Photonics Kraken Test Laser Technologie (HPE Confidential)
The question is how this new computing will be used. Will we see it in consumer devices, for instance? Answer: possibly, eventually. HPE hopes its memory-driven architecture will scale "easily" to an exabyte system and then beyond that to a "near-limitless" - and unimaginable - "pool of memory - 4,096 yottabytes." That is, says HPE, 250,000 times the entire digital universe today.
In any case, the company says memory-driven computing will find applications in Mars exploration, healthcare, personalized retail, and "a future without delays" in transportation. Well, bring it on (if that goes for public transport too)!
China currently holds the two fastest supercomputers. Earlier in 2017, officials there said the country was on track to unveil the world's first exascale machine,the Tianhe-3, by the year 2020. It would be able to handle - an equally unimaginable - one quintillion calculations per second.
But no matter who is working on what, they all face a mountain to climb - not least in creating the technology - but also on bringing down the cost.