Working with data is a core practice for research. The scale of research data sets, however, has increased many orders of magnitude over the past decade. UAB researchers have increasingly turned to resources provided by UAB IT Research Computing for their data analysis and management needs.  

Research Computing has seen a 600 percent increase in unique accounts since 2018. As research and technology evolve, so do the services offered by Research Computing. An important project is underway that positions Research Computing and UAB research for growth — key for UAB’s Growth with Purpose initiative to raise $1 billion in research funding.

UAB IT’s Research Computing team has been hard at work over the last year to bring updates to a multitude of areas, overseen in large part by John-Paul Robinson, manager and architect of high-performance computing, and his team.

“We’re excited to see these ideas come to fruition,” Robinson said. “These updates will help us support growth and improvements over the coming years. We have two major improvements that are already in the works.”

The first improvement is an ongoing data center migration for the compute capacity provided by Research Computing. The campus Technology Innovation Center (TIC) has been home to UAB’s primary Cheaha HPC compute fabric, 160+ servers that provide nearly 11,000 CPU cores, and the GPFS parallel file system. Together these resources help analyze and store much of UAB’s research data. To support growth, the compute resources are being moved to a new HPC-focused data hall at the Birmingham DC BLOX data center, where UAB IT is deploying a new version of the GPFS parallel file system. DC BLOX offers Tier III data center services in a number of locations across the Southeast. The HPC-centric data hall in Birmingham came online in December 2023. It provides dense power and updated cooling needed to support the demands of data analysis and AI applications. This move will allow Research Computing  to further expand storage capacity at the TIC data center in the future.

In addition to the move from TIC, Research Computing is consolidating compute resources that have already been located in traditional retail space at DC BLOX into this new HPC-focused data hall. This includes all the GPU resources currently offered by Research Computing for advanced computing, machine learning, AI workloads, and the platform services that support virtual machine (via OpenStack) and container (via Kubernetes) application modalities.

“Once we have this move complete, we’ll be able to see improvements in storage capacity and application performance,” Robinson said.

Research Computing’s upgrades to Cheaha’s HPC file storage are designed to improve performance and add capacity. Currently, the supercomputer is running on IBM’s GPFS 4. Robinson and his team are moving towards implementing the fifth generation of GPFS, along with upgrades to the hardware from Dell and Kalray. 

“The new GPFS version 5 file system implementation can supply data at 80 gigabytes per second with 160 gigabytes per second performance on the horizon with future upgrades,” Robinson said.

These moves are not without their challenges. Research Computing currently hosts more than 6 Petabytes of UAB’s research data.  Six petabytes is 6,000 Terabytes, which is the storage capacity equivalent of about 6,000 laptops. Migrating to a new file system means copying that data from the original location to the new location. 

Research Computing is hard at work designing the migration plans for the data and compute moves to minimize interruptions to UAB’s daily research computing workloads. 

“We will provide further details as the plans solidify and contact users directly to coordinate migrations with them,” Robinson said. “Our goal is to complete the data migrations mid-summer but may adjust the schedule as details are refined.”

As the summer progresses, the team will be adding on to their updates list. Improvements planned for coming months will be:

• Designing a new Research Computing Services network fabric
• Upgrading the ScienceDMZ with new hardware and network availability
• Consolidating Cheaha HPC, Cloud.rc, and k8s.rc at DC BLOX.

To learn more about the Research Computing team and the services provided to campus, visit UAB IT’s website.