Hi,

We have GPU enabled Ubuntu server (18.04) that is being used by multiple users simultaneously. Out requirement is to have utilization report generated for the server reporting how much each user has used the system (including GPU usage) . We are looking to tools but could not find suitable one. Please help.

Sysdig can help, but your 'requirement' is fairly vague. You don't say if it's real-time, aggregate over what period, spot-checks, etc., on what kind of hardware, or what you actually want to do with this data.

After being here for TWELVE YEARS, you should be aware that you need to post details and provide us with things you've done/tried/researched on your own, to avoid people giving you things you've already looked at. You've posted numerous threads over the years where you haven't responded or answered questions.

Hi,

Thank you for the help.

Let me brief on the requirement. It is not real time. It is per month utilization report we have to generate showing the cpu and gpu utilization by each user and overall system utilization. Spot checks would be an added advantage, but is not the main requirement. The hardware is Intel haswell processors (40 cpus), 256 GB RAM, 2 GPUs (V100). Hope this clarifies the requirement.


I thought the provided details might suffice. I have tried with atop, htop, collectl, xymon, but certainly it was not satisfying the requirement.

Taking your comments positively, sometimes I have might have missed something, but that was certainly not intentional. However, I would respond to the threads in future.


Thank you once gain.

Again, think about what you're asking. "Per month utilization report"....define that. Because that can be interpreted as:

• Checking user load ONCE PER MONTH.
• Checking user load DAILY and REPORTING once per month on the aggregate.
• Checking user load X times daily...
• ...etc., etc., etc..

Until you quantify what you need/want, your question remains vague. And how are you STORING the data, assuming you're capturing it per user, once per day?? Round robin graph? Database? CSV File? Again, too vague and too many variables. Again, after twelve years, you should know that any reporting task requires the same things:

• What to check
• How to check it in a meaningful way
• How often to check it
• What to DO with the data after you've checked

Define those things and your 'requirement' will be solved. You mentioned some tools above, but again saying it 'was not satisfying the requirement' doesn't say WHY it doesn't, so without knowing what problem you're having, there's no way anyone else can solve it.