INCLINE Resources¶

Allocations¶

Allocation Type	Description
`REAS Account`	Allocation for activities related to research projects described here. Granted on a quarterly basis based on competitive proposal process.
`CRSE Account`	Allocation for course-related activities. Allocations granted on a quarterly basis based on competitive proposal.
`MISC Account`	Limited “seed” accounts given on a first-come-first-served basis. Limits predetermined by committee based on demand and resources.
`EXTR Account`	Accounts for all off-campus and RMACC users. Granted by competitive proposal. Limits determined by committee.

Queues¶

Queue Name	Description
`compute`	Standard workhorse queue for jobs that use the compute nodes.
	12 Hours	10 Jobs
`compute-dev`	High-priority queue with tight limits on job length and queue number. Use for testing code but not for production runs.
	1 Hour	1 Job
`compute-long`	Low-priority queue with unlimited job length. Only one job allowed in the queue at a time
	Unlimited	1 Job
`gpu`	Queue for running jobs on the GPU nodes
`hm`	Queue for running jobs on the high memory nodes
`hetero`	Special queue for jobs that use any combination of compute, GPU, and high membory nodes.

Calculating Computing Units¶

INCLINE resources are described in terms of CPU Hours (CPUH). One standard CPUH is defined as one hour of allocated use by a compute node. Important: the CPUH will be billed to your account regardless of whether the node is fully utilized.

Time allocated on specialty nodes are denoted differently. One hour of allocated time by a gpu node is a GPU-CPUH; an hour of allocated time by a high memory node is a HM-CPUH. Resources for CPUH, GPU-CPUH, and HM-CPUH are billed independently, and are non-transferrable.

Example 1

Alice needs to run 10 large CFD simulations. She estimates that, running with approximately 2000 MPI processes (no special memory requirements), each simulation will complete in four hours.

To run a concurrent job with 2000 MPI processes will require 8 nodes. (8 x 256 threads = 2048). Each job will then require (4 hours) x (8 nodes) = 32CPUH. Therefore, Alice should request at least 320CPUH.

Example 2

Bob is working on a fictitious machine learning project that will require extensive data processing. He estimates that training and testing his model will take 2 days on a single A100 GPU. (There is no way to split the job across multiple GPUs; it must be done serially.) Afterwards, the data analysis will take about 30 hours on a high memory node.

Even though Bob will only be using one of the two GPUs on a gpu node, he must request time on the entire node. Because he also needs to use a high memory node, he must specify that allocation independently from the gpu node. Therefore, he should request 48 CPU-CPUH and 30 HM-CPUH.