On the theory that many heads in the cloud are better than one, three tech giants are giving some lucky researchers on-demand access to all the resources they need to work advancing cloud computing.
The HP, Intel and Yahoo Cloud Computing Test Bed, announced last week, is meant to promote open, collaborative research in cloud computing by providing large-scale environments where hardware and software designers can try out their large-scale ideas. The three giants funding the program are banking on smart people in academia, government, and industry to help come up with the best ways to build, connect, manage and deploy systems that deliver compute power on demand.
The test bed will be made up of six datacenters, hosted at HP Labs, Intel Research, Yahoo, the University of Illinois at Urbana-Champaign, Karlsruhe Institute of Technology in Germany, and the Infocomm Development Authority in Singapore. Each facility will house 1,000 to 4,000 processor cores (mainly HP and Intel hardware, naturally) running under the open-source distributed computing system Hadoop.
At the announcement of the new program, Prith Banerjee, HP’s senior vice president of research and director of HP Labs, said that enabling researchers to work on “intelligent infrastructure, radically new architectures, massive storage, and sustainable IT systems” will usher in the “era of ‘everything as a service’ — everything will act seamlessly through the cloud based on user location, preference, and community.” Banerjee said the project is essentially giving researchers their own “utility” to develop the cloud “as a platform for new services and experiences.”
“The test bed simulates a real-life, global, Internet-scale environment, which gives researchers an unprecedented ability to test applications and measure the performance of infrastructures and services built to run on large-scale cloud systems,” Russ Daniels, chief technology officer of HP Cloud Services, told GRIDtoday. “Because the test bed operates at Internet-scale, researchers can tackle larger problems, increase the speed at which experiments can be performed, investigate new parallel programming models, and train students that are at the frontier of the computing industry.”
Each of the big three will be contributing research results, as well. HP has five areas it’s focusing on. One, Daniels says, is to develop “an enterprise storage service that can automatically scale to store petabytes of data across geographically dispersed locations, and offers a non-stop, fault-tolerant, self-managing platform that can be simultaneously used by multiple parties for different purposes.”
Cells as a Service is a prototype system for managing cloud infrastructure based on “service cells.” Each cell can contain “an arbitrary assembly of virtual machines, virtual storage volumes, and virtual networks, connected in whatever design of IT infrastructure a customer desires,” Daniels says. “Cells are securely isolated from one another, although connectivity between cells can be established in controlled ways. This enables the management of a cloud infrastructure to be automated and allows for the dynamic deployment and management of services offered through the cloud.”
Another group will be working on ways to lower the energy consumption of the huge datacenters that will be required for cloud computing, Daniels says, while another focuses on low-cost “building-block components” that would reduce the cost of datacenters.
Intel is less specific about its cloud-specific research, but considering it makes the compute cores, it’s easy to imagine the company implementing functions that would enhance scalability and virtualization, or reduce heat, for example. This program falls under Intel’s “exploratory research” bucket, a spokesperson says, “which means it’s off-roadmap and more looking at the next five to 10 years.”
Hadoop! Pig! Real-World Experience!
Despite its trademark exclamation point (here omitted), Yahoo is no stranger to heavy computing research. After all, consider the business it’s in: serving millions of people who want their Web pages or mailboxes — pronto. The company launched its own co-op program for cloud-related research last year with the M45 Supercomputer Project, featuring a 4,500-processor cluster available for academics (Carnegie Mellon students being the first). The system, like the new test bed, runs on the Hadoop distributed computing/storage platform, which, along with the Pig programming environment, is one of the favorite subjects of Yahoo researchers. The company recently opened another supercomputer center in India, also doing cloud research.
“The Hadoop group in Yahoo is the key contributor to the open-source Hadoop software,” says Raghu Ramakrishnan, chief scientist for audience and research fellow with Yahoo Research (noting that Hadoop also powers IBM’s Blue Cloud and the IBM/Google cloud initiative). “Our current efforts are focused on improvements to Hadoop across the board to security, performance, and availability, as well as new features for richer workflow support, etc.”
Other major projects address “a number of cloud computing issues. For example: scalable, highly available, and globally replicated data serving and database services; scalable large-object stores; vertically partitioned databases; and machine virtualization.”
Yahoo brings something from its day-to-day business, as well, Ramakrishnan says. “In addition to the massively scalable infrastructure that we have developed and long deployed to operate our own Web applications, one of our core competences is building and managing some of the largest app services (for example, Mail) on the Web, and this constantly focuses and drives our research initiatives.”
Ramakrishnan says some of the biggest challenges facing cloud computing, and researchers, are immature management infrastructure, lack of a “coherent suite of cloud services from which entire applications can be built,” standard interfaces, and a need for “assurances of security and quality of service.”
Not All Blue Skies
The group says all the facilities should be up and running by the end of this year. Each center, “based on some common criteria, will determine who can and cannot become a member and whether the research work done on the platform is acceptable and legal,” HP’s Daniels says. “Additionally, there will be a steering committee, with representatives from the founding members and the other partners, that will determine who can and cannot become a new partner and host a Center of Excellence.”
Representatives of the big trio behind the effort are anticipating results to start showing up relatively soon after full launch, and the intended beneficiaries won’t be just people who read research papers.
“Businesses will ultimately benefit from the advancements that are made in cloud computing as a result of the test bed’s research findings in areas such as applications, software, infrastructure, datacenter management and so on,” Daniels says.
“Our hope,” Ramakrishnan says, “is that the concrete advances we’ve seen already in some of our projects will improve cloud services. Some of the available technologies are not as mature as we need them to be. We are making a sea change. As we shift to the service model, there will be rough spots.”