Flexible, scientific infrastructure for research on the future of cloud computing. Researchers use CloudLab to build their own clouds, experimenting with new architectures that will form the basis for the next generation of computing platforms.
CloudLab provides researchers with control and visibility all the way down to the bare metal. Provisioning an entire cloud inside of CloudLab takes only minutes. Most CloudLab resources provide hard isolation from other users, so it can support hundreds of simultaneous "slices", with each getting an artifact-free environment suitable for scientific experimentation with new cloud architectures. Run standard cloud software stacks such as OpenStack and Hadoop. Or, build your own from the ground up. The bare metal's the limit!
CloudLab is built from the software technologies that make up Emulab and parts of GENI, so it provides a familiar, consistent interface for researchers.
The CloudLab clusters have almost 15,000 cores distributed across three sites around the United States: Utah, Wisconsin, and South Carolina. Each cluster has a different focus: storage and networking (using hardware from Cisco, Seagate, and HP), high-memory computing (Dell), and energy-efficient computing (HP). CloudLab is interconnected with nationwide and international infrastructure from Internet2, so it is possible to extend private, software-defined networks right to every server.
When you build a cloud on CloudLab, you get a slice of the facility. In that slice, you have full control. This means you can run a full suite of cloud software of your own—compute, networking, and storage. That suite might look like one of today's cloud software stacks (for example, maybe it's an instance of OpenStack), it might be an incremental improvement to today's stacks (for example, replacing the storage layer), or it might be something radically different, built from the ground-up to support features like real-time computing, integration with cyber-physical systems, high performance computing, or energy awareness. Your cloud might be only for your own use as you experiment with your new architecture, or you could open it up to other users to get real application workloads.
CloudLab is built around profiles. A profile is a description of everything needed to build a cloud: the physical hardware (servers, disks, switches) and the software needed to transform it into a particular type of cloud. A profile is fully packaged and automated: building a cloud from scratch may take only minutes, depending on its size and complexity. CloudLab will provide "stock" profiles for several popular cloud software stacks, but users can modify these or build their own from scratch.
We've built widely-used testbeds for the computer science research community for decades, including Emulab, parts of GENI, and Apt. We're also heavily involved in reaching out to the community of computational research through the ACI-REF initiative, next-generation applications with transformative public benefit through US-IGNITE, and through a partnership with HTCondor, CloudLab users will be able to "Opt-in" to compute jobs as real workloads if they wish.
We recommend that users of CloudLab join the cloudlab-users mailing list. The list has a searchable archive, and is a good place to direct questions that are of general interest to CloudLab's user community. For questions that are not of general-interest, such as questions about individual accounts or experiments, send mail to firstname.lastname@example.org.
General inquiries or comments about CloudLab (not support-related) can be directed to email@example.com
The Acceptable Use Policy for CloudLab can be found here.