A paper which shows the current state of heterogeneous, reconfigurable hardware resources in the SAVI testbed has been presented by Stuart Byma, Hadi Bannazadeh, Alberto Leon-Garcia, J. Gregory Steffan and Paul Chow from the University of Toronto (Toronto, ON, Canada) has been presented at the 9th International Conference on Testbeds and Research Infrastructures for the Development of Networks and Communities (TridentCom 2014). It introduced the different types of virtualized reconfigurable resources in the testbed of the Smart Applications on Virtual Infrastructure (SAVI) network.
What is SAVI’s purpose? The purpose of the SAVI network is to investigate future application platforms that rely on virtualized, flexible infrastructure. This infrastructure is able to deploy large-scale, distributed systems that utilize the resources (wireless and wired networks, computing, devices) to deliver applications. The SAVI testbed consists of several main components: Core data center nodes with traditional cloud computing resources (VMs, storage, network), Smart Edge nodes that complement traditional cloud resources with heterogeneous resources (bare-metal servers, FPGAs, GPUs), Access Nodes that provide wireless connectivity, the SAVI testbed network that interconnects all components, and a Control Center to orchestrate applications and experiments.
Researchers described in this paper how the heterogeneous resources in the Smart Edge nodes are enabled, and then presented the different types of reconfigurable hardware resources in the SAVI testbed, and how each is controlled and managed. They presented several use cases for these resources, showing how they allow researchers to run experiments and applications that were previously impossible, and how virtualized reconfigurable resources can easily outperform applications run in software on Virtual Machines (VMs).
And finally, what are the plans for the future? Researchers plan to continue adding to the number of physical FPGA resources in the system, and expand these resources to all Smart Edge nodes in the testbed. They will also continue exploring the concept of Virtualized FPGA Resources, to see how closely they can be fit within the cloud computing model.
You can find the full paper here.
