4 de jun. de 2010

Os limites dos datacenters

O artigo não é novo, mas vale a pena republicar neste blog por causa da urgência do assunto: a capacidade dos datacenters para armazernar dezenas de milhares de informações postadas na rede, como mensagens, arquivos, imagens e vídeos: 


  Wall Street Journal, 20 de março de 2010


Messages, files, images, videos – they're being madly accumulated by the hundreds of millions of people now embracing gadget-filled lifestyles and ways of working.
But the abundance is testing the limits of the datacenters that must move and store it all. Companies behind booming "Web 2.0" services, like social networking, video sharing and blogging, and "cloud" computing businesses, which rent computing power or provide software as a service, are struggling to cope with increasingly complex and costly networks of countless server computers.
We've basically reached breaking point," says Drue Reeves, director of datacenter research at Burton Group, a research firm in Midvale, Utah. Companies don't have the money to buy still more servers, hire the people to manage them, or build giant facilities to house, power and cool them.
Helping datacenters deal with these problems is the focus of a number of the technology companies on The Next Big Thing list. Each is coming at it a different way, but all aim to help datacenters work better, faster, cheaper. Several are focused on technologies that boost server efficiency, while others concentrate on networking improvements. And two interesting standouts tackle the data itself—trying to make order out of chaos and deliver more usable information.
Server inefficiency
Servers are often standing idle, consuming power and space while they wait for other processes to happen. Efficiency has improved significantly in recent years thanks to virtualization, or software that turns one physical machine into multiple virtual machines, yet it remains a problem.
A key issue has been the bottleneck created by data-storage systems. Fusion-io Inc., No. 2 on this list, tackles this with newfangled systems that, instead of traditional spinning disks, use flash memory, a technology that has no moving parts and is best known for holding songs in iPods and images in digital cameras. The speed of flash and the way Fusion-io incorporates it into servers lets servers work three to ten times harder, the company says.
Akorri Networks Inc., No. 43, tries to help companies get more out of the datacenter they have with better management tools. Its technology helps them manage virtualized servers together with network and storage systems.
Azul Systems Inc., No. 6, and Schooner Information Technology Inc., No. 34, build specialized servers for specific datacenter software programs, which they say perform significantly better than general-purpose servers.
Azul's servers run applications written in Java, a popular programming language used for sophisticated Web sites and online applications like stock-trading systems. Co-founder and CEO Scott Sellers says one Azul machine replaces 10 to 20 general-purpose servers.
Schooner builds specialized servers to run two free middleware applications popular in Web 2.0 and cloud datacenters: MySQL, an open-source database program, and Memcached, a Web site memory caching system. The appliances fit easily into their networks and supply ten times more computing power than a general-purpose server because they exploit new technologies like "multi-core" computer chips and flash memory. Schooner's approach is an alternative to virtualization, which co-founder John Busch says doesn't work well for database or caching tasks.
Improving networks
Other list companies are focused on the networking gear that moves data from machine to machine. This is key because datacenters need better quality, faster networks capable of handling their increasingly souped-up servers, Mr. Reeves says.
Force10 Networks Inc., No. 41, a fast-growing maker of Ethernet switches and routers, is supplier to many Web 2.0 companies that want faster, more reliable and more efficient datacenters. And Silver Peak Systems Inc., No. 20, helps these companies move data among their datacenters at rapid speed, easing what have been onerous backup and recovery jobs.
"We focused on the scale problem," says founder David Hughes, who became convinced years ago that data volume was going up faster than bandwidth, or the speed at which it could be moved. Indeed, datacenters, lacking the bandwidth to move large amounts of data between datacenters, often send physical backup tapes by courier instead. Silver Peak lets them ship the data digitally fast by leveraging certain optimization techniques and the power of multi-core chips.
Meanwhile, GroundWork Open Source Inc., No. 28, has focused on making increasingly complex networks easier, cheaper, faster to manage. Its product provides a single user interface and other foundational systems for using many popular—and free —open-source network-management tools.
Using the data flood
While datacenters aren't necessarily their customers, two companies on the list are looking to improve how vast stores of data are structured so the load can be used in more interesting and useful ways by businesses and consumers.

ParAccel Inc., No. 18, makes database software that businesses such as retailers and financial-services firms use to mobilize many servers to query their data 10 to 20 times faster than they could in the past.
"It's been very apparent that the amount of information out there that needs to be analyzed to make real-time decisions in companies is growing far faster than the traditional database architectures have been built for," says Barry Zane, ParAccel's founder. By tweaking how data is structured and changing the architecture, "we get a lot more performance out of smaller number of computers."
Metaweb Technologies Inc., No. 26, is restructuring data on the Web to make it more useful to both consumers and businesses. With the help of a community of collaborators, it has created 12 million "entities" for people, places and things, and mapped how they relate to each other, a structure it says offers a better way to find and use information online, compared to keyword searches. The company's first product, due in March, helps Web sites use relevant content from elsewhere on the Web on their own sites.

Nenhum comentário: