Virtual machines are available in three general frameworks:
  • Small machines with two processors, two gigabytes of RAM, and 15 gigabytes of data storage with a range of applications. These are available on-demand to Duke researchers through the VCL (Virtual Computing Lab) and VM-Manage services. VCL provides short terms access to virtual machines, and these machines do not persist; VM-Manage provides long-term use (in semester-long segments) and these can be renewed for longer periods. These machines are provided without charge, and can be ordered up by visiting the VCL or VM-Manage web pages.
  • In contrast to the VCL and VM-Manage machines, tailored machines can be created with much larger memory and more CPUs (with examples as large as 256 GB RAM and 22 CPUs) and can be configured to use storage that is available from OIT. The virtual machines also can be provisioned in the Protected Data Network to allow analysis of sensitive data. Virtual machines are widely used at Duke as dedicated computational servers, providing researchers easy access to computational resources that are well fit for their specific needs. These can be fitted with Windows Server or Red Hat Enterprise Linux operating systems, and can be more finely tailored than typical cloud vendors. Contact to get more information.
  • Within the context of Research Toolkits (, RAPID virtual machines are available in a pilot extending through Spring Semester 2016. RAPID VMs are quite flexible and are configurable in various custom sizes of RAM and CPU-cores. The pilot was announced at the Duke Research Computing Symposium in January 2016, and Mark McCahill, a lead developer on the Research Toolkits project, presented on the service. This presentation is available online.
It is just barely an oversimplification to say that Duke Research Computing treats cloud computing as just another source of computing cycles with strengths and weaknesses, risks and rewards. Of course the complication is that the “units” are not the usual metal boxes or neatly packed virtual machines but can be whole infrastructures, littered with meters for determining use and assigning cost. Researchers at Duke have been able to take advantage of cloud services’ “cloud credit” programs, and Research Computing staff have coordinated, written, and submitted their applications. Because of the complex billing for cloud services, Real World costs of cloud resources can vary widely from anticipated costs. To help make cloud computing more predictable, the cloud credit programs are a good first step for some researchers.
Strategic use of cloud services for research computing makes business sense at Duke, and we have begun to weave cloud services into Duke Research Computing offerings — often quite invisibly. In Spring 2016, for example, compute cycles from the cloud were pulled in to handle high demand and to ease strains of cluster maintenance. In many cases, using the cloud for compute cycles is a matter of doing the business math, and the day will soon come when the cloud is the “right place” for many scientists to conduct their work. Duke Research Computing helps to move research to the cloud for good competitive reasons. Not because the cloud is the new shiny thing.