Design and Implementation of the Workflow of an Academic Cloud

Abstract

In this work we discuss the design and implementation of an academic cloud service christened Baadal. Tailored for academic and research requirements, Baadal bridges the gap between a private cloud and the requirements of an institution where request patterns and infrastructure are quite different from commercial settings. For example, researchers typically run simulations requiring hundreds of Virtual Machines (VMs) all communicating through message-passing interfaces to solve complex problems. We describe our experience with designing and developing a cloud workflow to support such requirements. Our workflow is quite different from that provided by other commercial cloud vendors (which we found not suited to our requirements).

Another salient difference in academic computing infrastructure from commercial infrastructure is the physical resource availability. Often, a university has a small number of compute servers connected to shared SAN or NAS based storage. This may often not be enough to service the computation requirements of the whole university. Apart from this infrastructure, universities typically have a few hundred to a few thousand “workstations” which are commodity desktops with local disk-attached-storage. Most of these workstations remain grossly underutilized. Our cloud infrastructure utilizes this idle compute capacity to provide higher scalability for our cloud implementation.