Flux Research Group / School of Computing
Apt logo

Apt

Now online at aptlab.net!

Apt (the Adaptable Profile-Driven Testbed) is a new type of facility: a meta-testbed that is adaptable through "profiles” to support a wide range of computing-based research domains. Apt focuses on getting the infrastructure out of the way, giving researchers the ability to create testbed environments ("profiles") that are tailored to their own domains, and to share those environments with their colleagues. Apt targets both researchers in computer science and researchers from other compute-intensive fields.

Apt consists of a hardware foundation and a testbed control system.

The software controlling Apt is built on systems previously built by the Flux group including Emulab and GENI.  It will enable the physical hardware to be adapted to a broad range of research domains through the application of domain-specific profiles. Each profile will provide an environment that is tailored to a particular domain - users of the instrument will have the option of using existing profiles developing their own profiles to share with their colleagues, collaborators, and students. Persistent profiles will provide a shared pool of resources, much like today’s clouds and compute clusters. The key difference is that, because they run on top of a shared environment, resources can be repurposed between profiles as needed—for example, to support “burst” needs of time-sensitive science. On-demand profiles will allow researchers will be able to “package up” the environment necessary to run experiments, including software, datasets, and a description of the configuration of the underlying environment. With a single click, other researchers will be able to instantiate a fully fledged testbed profile on Apt, ready to begin conducting research.

The Apt hardware, housed at the University of Utah Downtown Data Center, will feature a flexible interconnect fabric that supports dynamic configuration as either 40Gb Ethernet or 56Gb InfiniBand. It will provide further flexibility by offering both centralized and node-local storage, and support for software-defined networking.

current people

Eric Eide
Eric Eide
Faculty
Jonathon Duerig
Jonathon Duerig
Research staff
Dmitry Duplyakin
Dmitry Duplyakin
Research staff
Mike Hibler
Mike Hibler
Research staff
Leigh Stoller
Leigh Stoller
Research staff
Kirk Webb
Kirk Webb
Research staff
Gary Wong
Gary Wong
Research staff