The installation of a second 100-gigabit-per-second (Gbps) core router introduces performant redundancy for the OLCF's global users community and helps maintain year-round network availability. Previously the OLCF network consisted of one router with a single 100 Gbps connection and a backup router with multiple 10 Gbps links.
"This network enhancement provides redundancy for maintenance, outages, and upgrades and will allow us to quickly bring in new technology with minimal to no disruption", stated OLCF high-performance computing (HPC) Linux systems engineer and task lead Daniel Pelfrey. "It's a very good situation for moving forward."
Daniel Pelfrey teamed with OLCF HPC network administrators Paul Newman and Benton Sparks to complete the core router installation.
One of the primary beneficiaries of 100 Gbps connectivity is OLCF users, who regularly move large datasets to and from the centre. These datasets typically range from a few gigabytes to a few terabytes, but the OLCF's network has the capacity to handle even larger datasets. This capability is especially critical for researchers in data-intensive domains such as cosmology and climate science who possess multi-petabyte datasets and shuttle data between multiple HPC centres.
"At full saturation, a 100 Gbps link can move a petabyte of data in about 27 hours", Daniel Pelfrey stated.
The core router installation marks the completion of a multiyear process that began in 2015 when the OLCF established its first 100 Gbps link - a significant speed-up from the multiple 10 Gbps links that existed previously. The establishment of 100 Gbps connectivity from both core routers ensures that no single point of failure exists between the OLCF and ESnet scientific community, which spans national laboratories, HPC centres, and world-class experimental facilities in the United States and Europe.
The arrangement also positions the laboratory to respond to researchers' future data needs, which might soon require aggregating multiple 100 Gbps links to work in parallel, Daniel Pelfrey said.
"If there is a use case for it, I think that's where we're headed", he stated.