Priority Access
What is priority access?
In short, priority access to HPC resources accelerates research on the HPC. There are two main benefits:
Shorter wait times
Priority access is sort of like TSA PreCheck at airports. Users with priority access get to move through the line (read: job queue) faster than general users. Meaning priority users' jobs start running faster.Longer run times
Users with priority access can run their jobs indefinitely. Meanwhile, the time limit for most general users is generally 12 hours, limiting certain analyses.
All access to resources is managed through the cluster’s job scheduler. With priority access, computational jobs are moved higher in the queuing system, and in most cases begin execution within twelve hours, depending upon other FairShare factors. Accordingly, priority users' jobs are usually competing mostly with other priority jobs rather than jobs from the full HPC userbase.
How do I purchase priority access?
Priority access to HPC resources is available under a “condo model,” where faculty are able to purchase semi-dedicated nodes which get made available to all users when their lab members are not using them.
Under the condo model, faculty researchers fund the capital equipment costs of individual compute nodes, while the university funds the operating costs of running these nodes for five years. Faculty who purchase compute nodes receive access to equivalent resources at a higher priority than other researchers.
To be clear, researchers are not purchasing consistent access to a specific physical node. They are purchasing access to a node’s worth of specific (but interchangeable) resources. The faculty can then designate others—such as their graduate students, postdoctoral researchers, etc.—to receive priority access to those resources.
How will my lab members access our node(s)?
Priority users will be given a custom QoS to access to priority resources. We use QoSs to govern who can access priority resources (a.k.a. Trackable RESources, or TRes). Please note that priority users will not be given separate partitions.
How much does priority access cost?
It varies depending on what you want. We have provided estimates below which are accurate as of February 2024.
Service Type | Unit/Term | Unit Cost | Minimum Purchase |
Purchase (CPU) | 1 server/5 yrs | $18,000 (cpu) | 1 server |
Purchase (GPU)
| 1 server/5 yrs (1-3 GPU) | $14,500, plus $2500/GPU | 1 server, plus 1 GPU |
Rental (CPU) | 1 server/1 mo | $300 | 1 month/server |
Rental (GPU)
| 1 server/1 mo (1-3 GPU) | $242, plus $42/GPU | 1 month/ server/GPU |
This includes ITS infrastructure and licensing costs. Actual cost may differ and will depend on a formal quote from Dell.
The current hardware specifications are as follows:
CPU Node - Dell R6525 | GPU Node - Dell R7525 |
---|---|
2x 64-core AMD Epyc processors (128 cores) 512GB RAM 480GB SSD 25Gb/s Ethernet 200Gb/s InfiniBand (storage interconnect) | 2x 32-core AMD Epyc processors (64 cores) 512GB RAM 480GB SSD 25Gb/s Ethernet 200Gb/s InfiniBand (storage interconnect) 1x or 3x NVIDIA A100 GPU accelerator |
It is worth noting that the market supply of GPUs is currently low. Accordingly. Dell may need to substitute a different card in place of the A100.
I’m convinced. How do I purchase priority access?
Interested faculty can get started by submitting a PI Equipment Request Form. Storrs HPC Admins will then request a quote from Dell. When the quote is ready, it will be forwarded to PIs for review.
Please note: Interested parties should be direct affiliates of UConn.
We do not allow external businesses to directly purchase equipment, but current UConn faculty may use grant funds from corporate sources.