Ticket 19845 - TRESBillingWeights applied on nodes shared by multiple partitions
Summary: TRESBillingWeights applied on nodes shared by multiple partitions
Status: OPEN
Alias: None
Product: Slurm
Classification: Unclassified
Component: Accounting (show other tickets)
Version: 23.02.7
Hardware: Linux Linux
: 6 - No support contract
Assignee: Jacob Jenson
QA Contact:
URL:
Depends on:
Blocks:
 
Reported: 2024-05-10 08:08 MDT by Xu
Modified: 2024-05-10 11:36 MDT (History)
0 users

See Also:
Site: -Other-
Alineos Sites: ---
Atos/Eviden Sites: ---
Confidential Site: ---
Coreweave sites: ---
Cray Sites: ---
DS9 clusters: ---
HPCnow Sites: ---
HPE Sites: ---
IBM Sites: ---
NOAA SIte: ---
NoveTech Sites: ---
Nvidia HWinf-CS Sites: ---
OCF Sites: ---
Recursion Pharma Sites: ---
SFW Sites: ---
SNIC sites: ---
Tzag Elita Sites: ---
Linux Distro: ---
Machine Name:
CLE Version:
Version Fixed:
Target Release: ---
DevPrio: ---
Emory-Cloud Sites: ---


Attachments

Note You need to log in before you can comment on or make changes to this ticket.
Description Xu 2024-05-10 08:08:10 MDT
Hi, 

We have nodes with different hardware in a single partition. We would like to apply different TRESBillingWeights to them according to the hardware. 

Our thought is to create a "top level" partition, e.g., standard, and two "nested" partitions, e.g., standard-cheap, standard-expensive, under the top level partition, i.e., overlapping them. We then give different TRESBillingWeights to these "nested" partitions. 

When a user submit a job simply requesting "top level" partition, their job will eventually land on either "nested" partition, and should be charged based on the "nested" partition's TRESBillingWeights.

However, this seems impossible right now. TRESBillingWeights are applied based on the partition the user request, in this case, the "top level" partition.

Is there a way to do what we want? Thanks, 

P.S. we plan to give the user the flexibility to pick which "nested" partition to run their job, by the means of "constraint". However, if they don't pick, they will be allocated "cheap" one first using node weight.
Comment 1 Xu 2024-05-10 11:36:23 MDT
To add to my question: our ultimate goal is to be able to charge, in the sense of fairshare, users differently based on the types of hardware, e.g., different CPU, GPU cost different Billing TRES, NOT just their quantity. We would like to achieve this without adding partitions from the user's perspective. To users, they can keep using the old partitions, but within each partition, different types of hardware they land on are charged differently.