ParallelCluster 2.10.1 and SLURM: What EC2 instance types are supported?

Question

The docs seems to say that you can use a t2.micro as a master_instance or a compute_resource, just for testing and playing around. But I get an an error in the /var/log/parallelcluster/slurm_resume log when I try to run a job using a t3a.xlarge. The compute node won't get started up:  
  
""  
\[slurm_plugin.common:add_instances_for_nodes] - ERROR - Encountered exception when launching instances for nodes (x2) \['compute-dy-t3axlarge-1', 'compute-dy-t3axlarge-2']: An error occurred (InvalidParameterCombination) when calling the RunInstances operation: Cluster placement groups are not supported by the 't3a.xlarge' instance type. Specify a supported instance type or change the placement group strategy, and try again.  
""  
  
I looked through the EC2 cluster placement groups documentation and it says while t2 types aren't supported, plenty of other ones including t3 and t3a are. Is there a docs that says what exact kinds of EC2 instance types are supported as parallelcluster compute nodes when using SLURM? Of course I know I can use M4 and M5 but I'm just thinking for testing to use a cheaper instance type.  
  
Edited by: ChrisTaylorUSC on Dec 30, 2020 11:42 AM

Accepted Answer

Hi,  
  
‘t3a.xlarge’ cannot be use here because it is a burstable performance instance.  Burstable performance instances are the only instance types that use credits for CPU usage.  
According to Cluster placement group rules and limitations in this doc:    Instances in a cluster placement group cannot be burstable performance instances.  
Here is a list of burstable performance instance types: https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/burstable-credits-baseline-concepts.html  
  
Let us know if it helps

ParallelCluster 2.10.1 and SLURM: What EC2 instance types are supported?

Relevant content