Summary
Computational chemistry software licenses are among the most expensive line items in a drug discovery budget. When those licenses run on cloud HPC, poor integration creates problems: idle licenses burn money, contention between researchers stalls work, and GPU-heavy notebook sessions run around the clock even when no one is using them.
It’s not just an IT problem. Every hour a computational chemist spends waiting on license contention or debugging failed jobs is an hour not spent advancing a drug candidate. For organizations racing to move therapies from discovery to the clinic, infrastructure friction directly slows time-to-market and delays treatments reaching the patients who need them.
BioTeam integrated Schrödinger, Cresset Engine Broker, and Posit Workbench (formerly RStudio Workbench) into Slurm-managed AWS ParallelCluster environments for a biotech-focused client. Each integration required a different technical approach. The result: researchers access the tools they need without contention, and expensive compute resources run only when the work demands it.
This work builds on the broader infrastructure engagement described in Schrödinger at Scale: Enterprise HPC Infrastructure, Deployed in Minutes.
Challenge
The client runs drug discovery pipelines that depend on commercial computational chemistry software. Each product has its own licensing model, resource requirements, and expectations for how researchers interact with it. Three specific integration challenges stood out:
Schrödinger license utilization. Enterprise Schrödinger licenses often represent hundreds of thousands of dollars in annual spending. Many organizations invest in more license capacity than their on-premises infrastructure can consume. Cloud HPC provides the elastic compute needed to fully utilize those licenses, but only if the scheduler is aware of token availability. Without that awareness, jobs compete blindly for licenses. Failed jobs retry and consume cluster resources while producing no results. Researchers lose time diagnosing license errors instead of doing science.
Cresset Engine Broker and Slurm. Cresset’s Engine Broker manages computational chemistry job distribution, but it was not designed for Slurm-based cloud HPC. Connecting the two required custom integration work. No established playbook existed for this pairing.
On-demand Posit Workbench with GPU and high-memory nodes. Researchers needed interactive RStudio notebook environments – delivered through Posit Workbench – backed by GPUs and large-memory instances. Running these nodes continuously is expensive. Relying on manual shutdown means nodes stay up long after sessions end, burning budget with no work to show for it. The environment needed to provision heavyweight compute only when a researcher started a session and release it as soon as they finish.
Approach
Schrödinger: License-aware Slurm scheduling. BioTeam wired Schrödinger license awareness directly into the Slurm scheduler, drawing on deep experience with FlexLM and custom Slurm scripts. The scheduler holds jobs in a queue until sufficient license tokens are free. This prevents failed job launches, reduces wasted compute cycles, and gives researchers predictable queue behavior. BioTeam also architected shared license servers across business units, enabling multiple research groups to draw from a common license pool without stepping on each other’s licenses. The full license server deployment is automated alongside the HPC cluster, so new environments inherit a working configuration from day one. For a detailed technical comparison of Schrödinger license integration on ParallelCluster versus AWS PCS, see PCS or ParallelCluster for Schrödinger Suite?
Cresset Engine Broker: Custom integration with Slurm on ParallelCluster. BioTeam built the integration between Cresset’s Engine Broker and Slurm from scratch. The Engine Broker handles Cresset’s own job distribution logic. BioTeam connected it to the Slurm scheduler so that Cresset workloads respect cluster resource limits, coexist with other applications on shared compute nodes, and scale elastically with ParallelCluster.
Posit Workbench: On-demand GPU and high-memory notebook sessions. BioTeam integrated Posit Workbench with the Slurm cluster so that interactive notebook sessions launch as Slurm jobs. When a researcher starts an RStudio session, Slurm provisions a GPU- or high-memory-capable node via ParallelCluster’s elastic scaling. When the session ends, the node terminates. Researchers get the compute they need without waiting for IT to provision machines. The client avoids paying for idle GPU instances. BioTeam has delivered this same Posit Workbench integration pattern across multiple client engagements.
Outcomes
- Schrödinger license costs align with actual usage. License-aware scheduling eliminated blind job failures and reduced wasted compute hours tied to license contention.
- Cresset workloads run natively on cloud HPC. The Engine Broker integration gives researchers access to Cresset tools within the same Slurm environment they use for everything else.
- Interactive notebook costs scale to zero when idle. GPU and high-memory Posit Workbench sessions provision on demand and release automatically.
- Researchers spend time on science, not infrastructure. Automated provisioning and license-aware scheduling removed the manual troubleshooting that pulled computational chemists away from their pipelines.
- Integration risk is low from day one. BioTeam has delivered Schrödinger and Posit Workbench integrations across multiple engagements. Clients inherit battle-tested configurations.
