doc/html/fair_tree.shtml

<!--#include virtual="header.txt"-->

<!--

Fair Tree contributed by Brigham Young University
Authors: Ryan Cox and Levi Morrison

 -->

<h1>Fair Tree Fairshare Algorithm</h1>

<h2 id="contents">Contents<a class="slurm_link" href="#contents"></a></h2>
<ul>
<li><a href="#intro">Introduction</a></li>
<li><a href="#enduser">Overview for End Users</a></li>
<li><a href="#algorithm">Algorithm</a></li>
<li><a href="#fairshare">Level Fairshare Calculation</a></li>
<li><a href="#ties">Ties</a></li>
<li><a href="#sshare">sshare</a></li>
<li><a href="#config">Configuration</a></li>
<li><a href="#notes">Important notes</a></li>
</ul>

<h2 id="intro">Introduction<a class="slurm_link" href="#intro"></a></h2>

<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.</p>

<p>Some of the benefits include:</p>
<ul>
	<li>
		All users from a higher priority account receive a higher fair
		share factor than all users from a lower priority account.
	</li>
	<li>
		Users are sorted and ranked to prevent errors due to precision
		loss. Ties are allowed.
	</li>
	<li>
		Account coordinators cannot accidentally harm the priority of
		their users relative to users in other accounts.
	</li>
	<li>
		Users are extremely unlikely to have exactly the same fairshare
		factor as another user due to loss of precision in calculations.
	</li>
	<li>
		New jobs are immediately assigned a priority.
	</li>
</ul>

<h2 id="enduser">Overview for End Users
<a class="slurm_link" href="#enduser"></a>
</h2>

<p>This section is intended for non-admin users who just want to know how their
fairshare factor is determined.  Run <code>sshare -l</code> (lowercase "L") to
view the following columns: <code>FairShare, Level FS</code>.  Note that
Level FS values are infinity if the association has no usage.</p>

<p>If an account has a higher Level FS value than any other sibling user or
sibling account, all children of that account will have a higher FairShare value
than the children of the other account. This is true at every level of the
association tree.</p>

<p>The FairShare value is obtained by using the Fair Tree
<a href="#algorithm">algorithm</a> to rank all users in the order that they
should be prioritized (descending). The FairShare value is the user's rank
divided by the total number of user associations. The highest ranked user
receives a 1.0 fairshare value.</p>

<p>If you (UserA) have a lower FairShare value than another user (UserB) and
want to know why, find the first common ancestor account. At the level
below the common ancestor, compare the Level FS value of your ancestor to the
Level FS value of UserB's ancestor. Your ancestor has a lower Level FS value
than UserB's ancestor. For information on how Level FS value is
calculated, read the section about the <a href="#fairshare">Level FS
equation</a>.</p>

<p>For example, assume the association tree contains UserA and UserB as
follows:</p>
<pre>
root =&gt; Acct1 =&gt; Acct12 =&gt; UserA
root =&gt; Acct1 =&gt; Acct16 =&gt; UserB
</pre>

<p>Acct1 is the first common ancestor of UserA and UserB. Check the Level FS
values of Acct12 and Acct16.  If UserB has a higher FairShare value than UserA,
Acct16 has a higher Level FS value than Acct12.</p>

<p>The sections below contain more information about the algorithm, including
how the final fairshare factor and the Level FS values are calculated.</p>

<h2 id="algorithm">Algorithm<a class="slurm_link" href="#algorithm"></a></h2>

<p>An equation is used to calculate a Level Fairshare value for each
association, only considering the shares and usage of itself and its siblings.
A <a href="http://www.math.ucsd.edu/~ebender/CombText/ch-9.pdf">
rooted plane tree</a> <small>(PDF download)</small>, also known as a rooted
ordered tree, is logically created then sorted by Level Fairshare with the
highest values on the left. The tree is then visited in a depth-first
traversal. Users are ranked in pre-order as they are found. The ranking is
used to create the final fairshare factor for the user.</p>

<p>The algorithm performs a single traversal of the tree since all the steps
can be combined. The basic idea is to set <i>rank</i> equal to the count of user
associations then start at root:</p>

<ul><li>Calculate Level Fairshare for the subtree's children</li>
<li>Sort children of the subtree</li>
<li>Visit the children in descending order</li>
<ul><li>If user, assign a final fairshare factor similar to
	(rank-- / user_assoc_count)</li>
<li>If account, descend to account</li>
</ul></ul>


<h2 id="fairshare">Level Fairshare Calculation
<a class="slurm_link" href="#fairshare"></a>
</h2>

<p>The Level Fairshare equation is described below. Under-served associations
will have a value greater than 1.0. Over-served associations will have a value
between 0.0 and 1.0.
</p>

<pre>
LF = S / U
</pre>

<dl>
<dt>LF</dt>
<dd> is the association's Level Fairshare</dd>
<dt> S</dt>
<dd> also known as Shares Norm, S is the association's assigned shares
normalized to the shares assigned to itself and its siblings:
<nobr><code>S = Sraw<sub>self</sub> / Sraw<sub>self+siblings</sub></code></nobr>
</dd>
<dt> U</dt>
<dd> also known as Effective Usage, U is the association's usage normalized to
the account's usage:
<nobr><code>U = Uraw<sub>self</sub> / Uraw<sub>self+siblings</sub></code></nobr>
</dd>
</dl>

<p>U and S are in the range <nobr><code>0.0 .. 1.0</code></nobr>. LF is in the
range <nobr><code>0.0 .. infinity</code>.</nobr></p>

<h2 id="ties">Ties<a class="slurm_link" href="#ties"></a></h2>

<p>Ties are handled as follows:
<ul>
<li>Sibling users with the same Level Fairshare receive the same rank</li>
<li>A user with the same Level Fairshare as a sibling account will receive the
same rank as its highest ranked user</li>
<li>Sibling accounts with the same Level Fairshare have their children lists
merged before descending</li>
</ul>
</p>

<h2 id="sshare">sshare<a class="slurm_link" href="#sshare"></a></h2>

<p>sshare was modified to show the Level Fairshare value as <code>Level FS</code> when
the <code>-l</code> (long) parameter is specified.  The field shows the value for each
association, thus allowing users to see the results of the fairshare
calculation at each level.</p>

<p>Note: Norm Usage is not used by Fair Tree but is still displayed.</p>

<h2 id="config">Configuration<a class="slurm_link" href="#config"></a></h2>

<p>The following slurm.conf parameters are used to
configure the Fair Tree algorithm.  See slurm.conf(5) man page for more
details.</p>

<dl>
<dt>PriorityType</dt>
<dd>Set this value to "priority/multifactor".</dd>
<dt>PriorityCalcPeriod</dt>
<dd>PriorityCalcPeriod is the frequency in minutes that job half-life decay
and Fair Tree calculations are performed.</dd>
</dl>

<h2 id="notes">Important Notes<a class="slurm_link" href="#notes"></a></h2>
<ul>
<li>As the Fair Tree algorithm ranks all users, active or not, the
administrator must carefully consider how to apply other priority weights
in the priority/multifactor plugin. The <i>PriorityWeightFairshare</i> can be
usefully set to a much smaller value than usual, possibly as low as 1 or 2 times
the number of user associations.
</li>
<li>Fair Tree requires the <a href="accounting.html">Slurm Accounting
Database</a> to provide usage information and the assigned shares values.
</li>
<li><i>scontrol reconfigure</i> does not cause the Fair Tree algorithm to
run immediately, even if switching from a different algorithm. You may have to
wait until the next iteration as defined by <i>PriorityCalcPeriod</i>.
</li>
</ul>


<!-- -------------------------------------------------------------------- -->
<p style="text-align:center;">Last modified 26 June 2023</p>

<!--#include virtual="footer.txt"-->