<div dir="ltr"><span style="text-align:left;color:rgb(34,34,34);text-transform:none;text-indent:0px;letter-spacing:normal;font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant:normal;font-weight:400;text-decoration:none;word-spacing:0px;display:inline;white-space:normal;float:none;background-color:transparent">> However, I do think Scott's approach is potentially very useful, by directing </span><br style="margin:0px;text-align:left;color:rgb(34,34,34);text-transform:none;text-indent:0px;letter-spacing:normal;padding-bottom:0px;font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant:normal;font-weight:400;text-decoration:none;word-spacing:0px;white-space:normal;direction:ltr;background-color:transparent"><span style="text-align:left;color:rgb(34,34,34);text-transform:none;text-indent:0px;letter-spacing:normal;font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant:normal;font-weight:400;text-decoration:none;word-spacing:0px;display:inline;white-space:normal;float:none;background-color:transparent">> jobs < full node to one end of a list of nodes and jobs that want full nodes </span><br style="margin:0px;text-align:left;color:rgb(34,34,34);text-transform:none;text-indent:0px;letter-spacing:normal;padding-bottom:0px;font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant:normal;font-weight:400;text-decoration:none;word-spacing:0px;white-space:normal;direction:ltr;background-color:transparent"><span style="text-align:left;color:rgb(34,34,34);text-transform:none;text-indent:0px;letter-spacing:normal;font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant:normal;font-weight:400;text-decoration:none;word-spacing:0px;display:inline;white-space:normal;float:none;background-color:transparent">> to the other end of the list (especially if you use the partition idea to </span><br style="margin:0px;text-align:left;color:rgb(34,34,34);text-transform:none;text-indent:0px;letter-spacing:normal;padding-bottom:0px;font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant:normal;font-weight:400;text-decoration:none;word-spacing:0px;white-space:normal;direction:ltr;background-color:transparent"><span style="text-align:left;color:rgb(34,34,34);text-transform:none;text-indent:0px;letter-spacing:normal;font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant:normal;font-weight:400;text-decoration:none;word-spacing:0px;display:inline;white-space:normal;float:none;background-color:transparent">> ensure that not all nodes are accessible to small jobs).</span><br style="margin:0px;text-align:left;color:rgb(34,34,34);text-transform:none;text-indent:0px;letter-spacing:normal;padding-bottom:0px;font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant:normal;font-weight:400;text-decoration:none;word-spacing:0px;white-space:normal;direction:ltr;background-color:transparent"><div><br></div><div>Yes, not First In Last Out scheduling, more like</div><div>Fragmentary Entry  Fractional  Incoming First Out  Full Unreserved for MPI</div><div>FEFIFOFUM</div><div><br></div><div>I shall get my coat on the way out. </div><div><br></div><div><br></div><div><br></div><div><br></div><div><br></div><div><br></div><div><br></div><div><br style="margin:0px;text-align:left;color:rgb(34,34,34);text-transform:none;text-indent:0px;letter-spacing:normal;padding-bottom:0px;font-family:arial,sans-serif;font-size:12.8px;font-style:normal;font-variant:normal;font-weight:400;text-decoration:none;word-spacing:0px;white-space:normal;direction:ltr;background-color:transparent"></div></div><div class="gmail_extra"><br><div class="gmail_quote">On 12 June 2018 at 06:33, Chris Samuel <span dir="ltr"><<a href="mailto:chris@csamuel.org" target="_blank">chris@csamuel.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi Prentice!<br>
<span><br>
On Tuesday, 12 June 2018 4:11:55 AM AEST Prentice Bisbal wrote:<br>
<br>
> I to make this work, I will be using job_submit.lua to apply this logic<br>
> and assign a job to a partition. If a user requests a specific partition<br>
> not in line with these specifications, job_submit.lua will reassign the<br>
> job to the appropriate QOS.<br>
<br>
</span>Yeah, that's very much like what we do for GPU jobs (redirect them to the <br>
partition with access to all cores, and ensure non-GPU jobs go to the <br>
partition with fewer cores) via the submit filter at present..<br>
<br>
I've already coded up something similar in Lua for our submit filter (that only <br>
affects my jobs for testing purposes) but I still need to handle memory <br>
correctly, in other words only pack jobs when the per-task memory request * <br>
tasks per node < node RAM (for now we'll let jobs where that's not the case go <br>
through to the keeper for Slurm to handle as now).<br>
<br>
However, I do think Scott's approach is potentially very useful, by directing <br>
jobs < full node to one end of a list of nodes and jobs that want full nodes <br>
to the other end of the list (especially if you use the partition idea to <br>
ensure that not all nodes are accessible to small jobs).<br>
<br>
cheers!<br>
<span class="im HOEnZb">Chris<br>
-- <br>
 Chris Samuel  :  <a href="http://www.csamuel.org/" target="_blank" rel="noreferrer">http://www.csamuel.org/</a>  :  Melbourne, VIC<br>
<br>
</span><div class="HOEnZb"><div class="h5">______________________________<wbr>_________________<br>
Beowulf mailing list, <a href="mailto:Beowulf@beowulf.org">Beowulf@beowulf.org</a> sponsored by Penguin Computing<br>
To change your subscription (digest mode or unsubscribe) visit <a href="http://www.beowulf.org/mailman/listinfo/beowulf" target="_blank" rel="noreferrer">http://www.beowulf.org/<wbr>mailman/listinfo/beowulf</a><br>
</div></div></blockquote></div><br></div>