High Throughput & Resilient Fabric Deployments on FermiCloud

Steven C. Timm, Fermi National Accelerator Laboratory

The FermiCloud Project has done considerable research on the use of virtualized Infiniband for MPI and also as a virtualized private network.  FermiCloud also hosts the test machines that participate in the US ESNet 100GBit/s WAN testbed.  We will discuss some of the use cases of our software-reconfigurable Infrastructure-as-a-Service facility thus far.

We will also present the resilient high-availability features of our network and SAN infrastructure which are designed to keep the facility operational in the event of the loss of one of our two buildings. We will present some of the lessons we have learned from grid computing and the early days of cloud computing and offer perspectives on future development directions that are needed.