Skip to content

Latest commit

 

History

History
131 lines (109 loc) · 6.33 KB

README.md

File metadata and controls

131 lines (109 loc) · 6.33 KB
Raw

seff-account

A modification of seff-array by Yale Center for Research Computing which is an extension of the Slurm command 'seff' designed to handle summarizing job usage for users and accounts over a period of time displays the information in a histogram. Note that if the user does not belong to the relevant account no data will be returned unless that user is an Operator or Administrator in the slurm database.

seff-account generates three types of histograms:

1. CPU Efficiency (utilization vs runtime)
1. Maximum memory usage versus the requested memory
2. Runtime of each job compared to the requested wall-time

Usage:

seff-account [-A account] [-c cluster] [-u username] [-S start time] [-E end time]

If run on a cluster that shares a single Slurm database, you can pass the name of the alternate cluster via -c cluster. The SLURM_CLUSTER_NAME env-var is checked and passed to sacct if present.

Start time and end time are formatted the same as the sacct command. By default it looks over the last day.

Understanding the output

Below is an example of jharvard_lab from start time September 15th, 2024 at 00:00:00 (midnight) to end time September 20th, 2024 at 00:00:00.

[jharvard@holylogin07 ~]$ seff-account -A jharvard_lab -S 2024-09-15T00:00:00 -E 2024-09-20T00:00:00
--------------------------------------------------------
Job Information
Names: unzip_systems.sh, interactive, process_plinder.sh, vscode, dummy4XDMod, .fasrcood/sys/dashboard/sys/Jupyter, install_env.sh
Cluster: odyssey
Users: user1, user2, user3, user4, user5, user6
Account: jharvard_lab
Start Time: 2024-09-15T00:00:00
End Time: 2024-09-20T00:00:00
Average Requested CPUs: 1.58 cores on 1.00 node(s)
Average Requested Memory: 18.82G
Average Requested Time: 16038.46s
--------------------------------------------------------
Job Status
CANCELLED by 98765: 1
COMPLETED: 8
FAILED: 5
OUT_OF_MEMORY: 1
TIMEOUT: 11
--------------------------------------------------------
--------------------------------------------------------
Job Statistics
Total Number of Jobs: 26
Average CPU Efficiency 7.98%
Average Memory Usage 2.87G
Average Run-time 9187.96s
---------------------

CPU Efficiency (%)
---------------------
+0.00e+00 - +1.00e+01  [18]  ████████████████████████████████████████
+1.00e+01 - +2.00e+01  [ 5]  ███████████▏
+2.00e+01 - +3.00e+01  [ 1]  ██▎
+3.00e+01 - +4.00e+01  [ 1]  ██▎
+4.00e+01 - +5.00e+01  [ 0]
+5.00e+01 - +6.00e+01  [ 1]  ██▎
+6.00e+01 - +7.00e+01  [ 0]
+7.00e+01 - +8.00e+01  [ 0]
+8.00e+01 - +9.00e+01  [ 0]
+9.00e+01 - +1.00e+02  [ 0]

Memory Efficiency (%)
---------------------
+0.00e+00 - +1.00e+01  [18]  ████████████████████████████████████████
+1.00e+01 - +2.00e+01  [ 1]  ██▎
+2.00e+01 - +3.00e+01  [ 2]  ████▌
+3.00e+01 - +4.00e+01  [ 1]  ██▎
+4.00e+01 - +5.00e+01  [ 1]  ██▎
+5.00e+01 - +6.00e+01  [ 0]
+6.00e+01 - +7.00e+01  [ 1]  ██▎
+7.00e+01 - +8.00e+01  [ 1]  ██▎
+8.00e+01 - +9.00e+01  [ 0]
+9.00e+01 - +1.00e+02  [ 1]  ██▎

Time Efficiency (%)
---------------------
+0.00e+00 - +1.00e+01  [11]  ████████████████████████████████████████
+1.00e+01 - +2.00e+01  [ 1]  ███▋
+2.00e+01 - +3.00e+01  [ 1]  ███▋
+3.00e+01 - +4.00e+01  [ 0]
+4.00e+01 - +5.00e+01  [ 0]
+5.00e+01 - +6.00e+01  [ 1]  ███▋
+6.00e+01 - +7.00e+01  [ 0]
+7.00e+01 - +8.00e+01  [ 0]
+8.00e+01 - +9.00e+01  [ 1]  ███▋
+9.00e+01 - +1.00e+02  [11]  ████████████████████████████████████████
--------------------------------------------------------
  • Job Information: summary of all jobs
    • Names: job names
    • Users: users from jharvard_lab that ran jobs between start time and end time
    • Account: slurm fairshare account/PI lab that was queried
    • Start Time: start time for the seff-account analysis
    • End time: end time for the seff-account analysis
    • Average [field]: average of all jobs between Start and End times
  • Job Status: indicates the total number of jobs on each state between Start and End Time
    • CANCELLED by 98765: total of job cancelled by user 98765 (you can check which uid is yours with the command id)
  • Finished Job Statistics: summary of all finished jobs. This includes completed, failed, out_of_memory, timeout, and preempted jobs. It excludes pending, running, and cancelled jobs.
  • CPU Efficiency (%): horizontal histogram representing the number of jobs that falls within that CPU % efficiency
  • Memory Efficiency (%): horizontal histogram representing the number of jobs that falls within that memory % efficiency
  • Time Efficiency (%): horizontal histogram representing the number of jobs that falls within that time % efficiency

Reading the histogram

Keep in mind that plotting a graph on the command line is not an easy task.

Each line represents a bin on a histogram. First line is from 0 to 10% CPU efficiency, second line is from 10 to 20% CPU efficiency, third line is from 20 to 30% CPU efficiency, and so on.

In this example, we can see that most jobs (18 out of 26) had a CPU efficiency of 10%. Thus, users in jharvard_lab, should investigate which jobs are not CPU efficient and then look into improving their CPU efficiency.

+0.00e+00 - +1.00e+01  [18]  ████████████████████████████████████████
+1.00e+01 - +2.00e+01  [ 5]  ███████████▏
+2.00e+01 - +3.00e+01  [ 1]  ██▎
+3.00e+01 - +4.00e+01  [ 1]  ██▎
+4.00e+01 - +5.00e+01  [ 0]
+5.00e+01 - +6.00e+01  [ 1]  ██▎
+6.00e+01 - +7.00e+01  [ 0]
+7.00e+01 - +8.00e+01  [ 0]
+8.00e+01 - +9.00e+01  [ 0]
+9.00e+01 - +1.00e+02  [ 0]

Improving your jobs

Ideally, most jobs would fall in to 90 to 100% CPU, memory, and time efficiency (last line of a histogram). For more information on improving your jobs, see Job Efficiency and Optimization Best Practices