The era of Big Data Computing is already here. Data centers today are reaching 1 million meter square each, while the power required to operate such center reaches close to 100 MWatts each. The electrical cost of such centers dominates operating expenses. In 2014, in the US alone, Data Centers energy consumption was at about 100 Billion KWh at a cost of close to $10B (at total of 15,000 Mega Watts).
The Power Usage Efficiency – PUE – measuring total energy of data centers vs. IT energy, ranges between 1.2 to 2.0. This means that one Joule saved in computing technology (IT), saves around 1.5 Joule of energy consumed by a data center.
Computing Technology energy is dominated by data movement and operations’ duration; our past research (N. Magen 2004) showed that >60% of the energy consumed in computing chips is due to data movement. Furthermore, Stanford research (B. Dally, M. Horowitz) shows that the energy ratio between execution (e.g. 64 bit DP op – 20pJ) and moving data (DRAM access; 256bit Read/Write -16nJ) is 2-3 order of magnitude. Furthermore, some of Big Data flows may creat Bandwidth bottlenecks which the Funnel concept may solve.
In this talk we will cover our previous research and disclose energy saving aspects; we will reveal a tip of new paradigm in handling data movements in Big Data environment.