w 26 | ocus cnrsI InternatIonal magazIne F Data torage:s scale would provide researchers with a A Teething Problem new way of addressing today’s healthcare challenges. Data. Its analysis needs to be handled by Tdata must be stored somewhere. huge amounts of data, we have seen theWeb pages. “Withkeep track of billions ofhe constantly generated flood ofDATAETED Cov It is easy to get lost in the depths of Big experts: data mining specialists, who need The obvious solution is to multiply advent of specific massively parallel solu- to bring to the surface the knowledge the number of storage units like hard tions which involve dividing the task so buried under the mass of digital informa- disks on computers or memory chips on that many machines address specific tion. “This difficult algorithmic work is mobile devices (see box). While this gen- parts,” explains Valduriez. similar to sifting through a draughtboard eral principle can apply to massive with billions of rows and columns to amounts of data, its implementation is BloomInG DATACEnTErs identify a handful of repeated pat- not yet well defined. “Think of storage as Datacenters housing these massively terns,” explains Jean-François Boulicaut, a construction site,” says Patrick parallel computing and storage systems researcher at the LIRIS3 in Villeurbanne. Valduriez, Inria senior researcher at the are cropping up all over the world. “This With his team, he is currently applying LIRMM.1 “Speeding up operations is particularly true in colder regions since this expertise to the analysis of large vol- means hiring more workers. But coordi- all these machines require continuous umes of urban and environmental data nating their activities also becomes more watercooling, which becomes expensive,” (the Amadouer project). This project ex- complex.” In other words, piling up stor- adds Valduriez. These centers are ac- plores the databases of the greater Lyon age systems is not enough: the way they cessed via a private network or the area to collect information on road traffic, work together must also be optimized. Internet. The system, known as “cloud energy, and pollution. Once analyzed, this This is all the more difficult as, in addition computing,” makes it possible to rent data may be used to devise, for instance, a to being plentiful, the data is heteroge- storage space or even computing time, new transportation policy for the city neous and dynamic. either temporarily or permanently (see center. There is however a downside: this Hence the emergence of new ways to figure below). This is the model used for type of information is of great interest to store information, notably driven by Amazon’s “Elastic Compute Cloud,” a the private sector, which would like to use Internet giants like Google which need to service mainly targeting businesses, or it for commercial purposes, sometimes at the expense of the scientific community. “Twenty years ago, a company wanted to buy the rights to the digital photos held by ClouD ComPuTInG the association of French museums,” notes Boulicaut. “As the quality of digital pictures could not rival that of film-based Storage photography at the time, the Ministry of Applications Culture almost agreed, before deciding against it.” Today’s politicians will hope- SupplierComputing server fully be as clear-sighted with respect to the Big Data gold rush. 01.laboratoired’informatiquedegrenoble(cnrs/ Internet or universitédegrenoble-I,-II,and-III/Institut polytechniquedegrenoble) private network 02.Informatique,signauxsystèmesdesophiaantipolis (cnrs/Inria/universitédenicesophiaantipolis). 03.laboratoired’informatiqueenimageetsystèmes d’information(cnrs/universitéclaudeBernard/ universitélyon-II/Insa/ecolecentraledelyon). Users contactInFormatIon: marie-Christinerousset > marie-christine.rousset@imag.fr Johanmontagnat Enterprise Personal Mobile > johan@i3s.unice.fr Computer/ Jean-François Boulicaut IF > jean-francois.boulicaut@insa-lyon.fr atag D ©
CIM28
To see the actual publication please follow the link above