Ever wonder what goes on behind the scenes at Research Computing (RC)? While most of the job involves long stretches at the computer terminal, every now and then we get to step out from behind the desk and journey to the server room. These chilled rooms hold all the data, storage, networking, and compute that make Odyssey so magical and powerful. The majority of RC's servers are located at MGHPCC and 1 Summer Street in Boston, but there are still some one-off server rooms kicking around Harvard campus.
Back in the day the model for high performance computing (HPC) was decentralized. Labs and departments had their own embedded mini-clusters. Instead of pooling resources in a centralized facility, compute was spread out across campus and managed by those who purchased the equipment. HPC silos were the norm. This all changed within the last decade as more people required access to HPC resources and cloud applications became ubiquitous. As the need for compute spread and the cloud expanded the silo model no longer made sense. Centralization and the condo model of computing were the path forward.
It is this path that brought some of team RC out from the cubicle on Monday. Brian White, Luis Silva, Henry Perkins, Mike Ethier, and Josue Rivera decommissioned the Converse server room. While the compute in Converse was online as part of the larger Odyssey cluster, the hardware was still sequestered. Over the course of a couple hours, Converse was broken down, packed up and shipped off to its new centralized home at 1 Summer Street. What exactly does a decommission look like? Check-out the below videos.