Skip to content
GitLab
Explore
Sign in
Primary navigation
Search or go to…
Project
S
Slurm
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Build
Pipelines
Jobs
Pipeline schedules
Artifacts
Deploy
Releases
Package Registry
Model registry
Operate
Environments
Terraform modules
Monitor
Incidents
Service Desk
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Terms and privacy
Keyboard shortcuts
?
Snippets
Groups
Projects
Show more breadcrumbs
tud-zih-energy
Slurm
Commits
8aa21bc1
Commit
8aa21bc1
authored
10 years ago
by
Morris Jette
Browse files
Options
Downloads
Plain Diff
Merge branch 'slurm-14.03'
parents
fe16627c
d036e332
No related branches found
No related tags found
No related merge requests found
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
doc/html/slurm_ug_agenda.shtml
+14
-13
14 additions, 13 deletions
doc/html/slurm_ug_agenda.shtml
with
14 additions
and
13 deletions
doc/html/slurm_ug_agenda.shtml
+
14
−
13
View file @
8aa21bc1
...
...
@@ -646,18 +646,19 @@ and involved in several research challenges of the European Human Brain Project.
<h3>Level-based job prioritization</h3>
<p>Ryan Cox and Levi Morrison (Brigham Young University)</p>
<p>We will present about our new LEVEL_BASED job prioritization mechanism.
The algorithm prioritizes users such that users in an under-served account
will always have a higher fair share factor than users in an over-served
account. It recurses through the account tree, calculates fair share at
each level, then uses bitwise math to ensure that the effect of usage and
shares below the current level cannot affect calculations at the current
level.</p>
<p>Basically, if accounts A and B have the same shares but A has higher usage
than B then children of account A will have a lower fair share factor than
children of account B. This is not guaranteed in other prioritization methods.
LEVEL_BASED was also designed to reduce the likelihood of errors due to
floating point precision loss.</p>
<p>We will begin by highlighting some problems with existing fairshare
algorithms. We will then present about our new Fair Tree algorithm which
is designed to avoid these issues.</p>
<p>Fair Tree prioritizes users such that if accounts A and B are siblings and
A has a higher fairshare factor than B, all children of A will have higher
fairshare factors than all children of B. The algorithm uses classic computer
science techniques to achieve this goal. First, the traditional fairshare
equation, 2**-(Usage/Shares), is used but it only considers an association
and its siblings, rather than including the effect of parents or children
in the equation. It then sorts the association tree by fairshare value to
create a rooted plane tree and uses a depth-first traversal to rank users
in the order they are found. This ranking is used to create the final
fairshare factor.</p>
<h3>Integrating Layouts Framework in Slurm</h3>
<p>Thomas Cadeau (BULL), Yiannis Georgiou (BULL), Matthieu Hautreux (CEA)</p>
...
...
@@ -865,6 +866,6 @@ and involved in several research challenges of the European Human Brain Project.
system to dynamically allocate disk space on our test high-IOPS SSD
scratch system.</p>
<p style="text-align:center;">Last modified
4
September 2014</p>
<p style="text-align:center;">Last modified
9
September 2014</p>
<!--#include virtual="footer.txt"-->
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment