<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<div class="moz-cite-prefix">On 13-06-17 12:36 AM, Laurie Kell
wrote:<br>
<br>
The SPC guys use Condor for running many instances of a program
(MULTIFAN) . They also use it for <br>
calculating the Hessian over multiple machines. From my point of
view these are legitimate, but trivial<br>
instances of parallelization. If you are interested you might
contact Nick Davies (when he surfaces).<br>
<br>
There are multi-threaded versions of the Blas which I have tested
for doing LU decomposition.<br>
These are very fast for large matrices compared to the current
code. The interesting thing is<br>
to understand their obtuse fortran formulations so that
corresponding adjoint code can be written<br>
in C++. I would be interested in collaborating on that.<br>
<br>
My current posting was more about true multi-threading, using
pthreads. <br>
I have a few trivial examples to show where the boilder plate goes
and <br>
an application to Multifan where we calculate the stuff for tag
returns for different tag releases<br>
in parallel as well as a bit more parallelization. <br>
<br>
Sibert also has done a "real world" example of this. His appears
to get much better results<br>
that I do. At present I can get Multifan to run in about 57% of
the time for multi-threaded version<br>
on an I7 laptop Sibert gets approx 5 fold improvement on some
multicore big machine.<br>
<br>
I also would like the extend the multi-threading by adding the
ability to transparently use multiple hosts.<br>
It two threads are on the same host they communicate via pthreads,
if on different hosts<br>
they communicate via MPI. this should be almost effortless for
the user.<br>
<br>
<br>
Dave<br>
<br>
<br>
<br>
</div>
<blockquote
cite="mid:CAAJ=99k1wZg97jB1+qLgsJsHtatEy+eZK5x+D+ezq6V_EgDkHA@mail.gmail.com"
type="cite">
<div dir="ltr">
<div>There are two main areas that we are looking at in the tuna
RFMOs and elsewhere for running ADMB and AutoDiff code i.e. <br>
i) running many instances of a single program e.g. over a grid
or ii) using parallel computing for a single operation such as
inverting a hessian. <br>
<br>
</div>
<div>Looking at some examples of each and then thinking about
ways forward could be a start.<br>
</div>
<div><br>
</div>
Laurie<br>
</div>
<div class="gmail_extra"><br>
<br>
<div class="gmail_quote">On Sat, Jun 15, 2013 at 2:35 AM, dave
fournier <span dir="ltr"><<a moz-do-not-send="true"
href="mailto:davef@otter-rsch.com" target="_blank">davef@otter-rsch.com</a>></span>
wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">On
13-06-14 01:01 PM, Laurie Kell wrote:<br>
<br>
<br>
I'm happy to collaborate, but how would it function. I
only want to do 90% of the work.
<div class="HOEnZb">
<div class="h5"><br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
multi-threading is something that I would like to be
involved in, and I know of others who would be keen to
collaborate.<br>
<br>
Laurie<br>
<br>
<br>
<br>
On 14/06/13 16:33, dave fournier wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
On 13-06-14 01:40 AM, Mark Payne wrote:<br>
<br>
<br>
I wish that all the players could be encouraged to
discuss future<br>
directions BEFORE the actual meeting.<br>
I think that is what the mail lists are for. There
is not enough time<br>
at a meeting to initiate a discussion on<br>
any substantive issue and develop it to a useful
level.<br>
<br>
For example at the recent meeting Sibert and I
presented some results on<br>
multi-threading which indicate<br>
that it can be useful enough for ADMB programs to
justify serious<br>
development efforts. Getting this together<br>
required about 3 months of programming/discussion to
prepare for the<br>
meeting.<br>
<br>
With regard to multi-threading, this looks
interesting.<br>
<br>
<a moz-do-not-send="true"
href="http://www.tomshardware.com/news/Intel-Xeon-Phi-Coprocessor-CPU,22700.html"
target="_blank">http://www.tomshardware.com/news/Intel-Xeon-Phi-Coprocessor-CPU,22700.html</a><br>
<br>
Dave<br>
<br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
Have you considered having a users session on the
side of the ASC - after all, there will be many of
the ICES users there.... Wednesday afternoon is
usually free for that sort of thing. And it might
be nice for some of us non-developers to chat with
you, hear future directions, and give feedback....<br>
<br>
Mark<br>
________________________________________<br>
Fra: <a moz-do-not-send="true"
href="mailto:users-bounces@admb-project.org"
target="_blank">users-bounces@admb-project.org</a>
[<a moz-do-not-send="true"
href="mailto:users-bounces@admb-project.org"
target="_blank">users-bounces@admb-project.org</a>]
på vegne af Arni Magnusson [<a
moz-do-not-send="true"
href="mailto:arnima@hafro.is" target="_blank">arnima@hafro.is</a>]<br>
Sendt: 13. juni 2013 23:07<br>
Til: <a moz-do-not-send="true"
href="mailto:developers@admb-project.org"
target="_blank">developers@admb-project.org</a>;
<a moz-do-not-send="true"
href="mailto:users@admb-project.org"
target="_blank">users@admb-project.org</a><br>
Cc: Athol Whitten; Rishi Sharma; Andre Punt; Dave
Fournier; Kasper Kristensen; Laurie Kell; Polina
Levontin<br>
Emne: [ADMB Users] ADMB meeting in Reykjavik -
Doodle Poll<br>
<br>
Dear friends,<br>
<br>
The upcoming ADMB Developers' workshop in
Reykjavik in September was<br>
discussed a bit at the ADMB Foundation board
meeting last week. There are<br>
two other relevant conferences taking place in
Reykjavik in the same<br>
month:<br>
<br>
12-14 Sep (Thu-Sat), LGM conference in
Reykjavik<br>
<a moz-do-not-send="true"
href="https://sites.google.com/site/lgm2013ice/"
target="_blank">https://sites.google.com/site/lgm2013ice/</a><br>
<br>
23-27 Sep (Mon-Fri), ICES conference in
Reykjavik<br>
<a moz-do-not-send="true"
href="http://ices.dk/news-and-events/asc/ASC-2013/Pages/default.aspx"
target="_blank">http://ices.dk/news-and-events/asc/ASC-2013/Pages/default.aspx</a><br>
<br>
Initially, we (Anders, Arni, Hans) scheduled the
ADMB meeting 8-11<br>
September, just prior to the LGM conference. Now
that our LGM abstract has<br>
been rejected, our focus is shifting to the ICES
conference.<br>
<br>
It seems likely that some ADMB-oriented people
might attend the ICES<br>
conference. To make it practical to visit Iceland
and attend both the ADMB<br>
and ICES events, we could shift the ADMB meeting
to be just before or<br>
after the ICES meeting.<br>
<br>
---<br>
<br>
Therefore, I have set up a Doodle Poll:<br>
<br>
<a moz-do-not-send="true"
href="http://doodle.com/dx5et3fe4dfp2d48"
target="_blank">http://doodle.com/dx5et3fe4dfp2d48</a><br>
<br>
Please mark the days that you would _prefer_. If
there are days you<br>
_cannot_ make it, please email me so I can take
that into account. Ignore<br>
this poll if you don't plan to attend. Feel free
to forward this email to<br>
others.<br>
<br>
At the end of the poll, we are looking for a 4-day
slot for the ADMB<br>
meeting.<br>
<br>
The range between 17 Sep and 1 Oct would be
practical for anyone planning<br>
to attend both the ADMB and ICES events. If I'm
the only one attending<br>
both, there may be no practical reason to shift
the ADMB meeting.<br>
<br>
Let's finish this poll before Wednesday (19 June)
and make a decision.<br>
<br>
Looking forward to welcoming you to Iceland this
autumn,<br>
<br>
Arni<br>
<br>
P.S. If you urgently need a reason to attend the
ICES conference, I can<br>
add you as a coauthor on a simple paper that's
been accepted in Session Q<br>
on Bayesian methods. Your contribution would be in
the introduction and<br>
discussion sections. This might help convince your
employer about the need<br>
to travel to Reykjavik this autumn.<br>
_______________________________________________<br>
Users mailing list<br>
<a moz-do-not-send="true"
href="mailto:Users@admb-project.org"
target="_blank">Users@admb-project.org</a><br>
<a moz-do-not-send="true"
href="http://lists.admb-project.org/mailman/listinfo/users"
target="_blank">http://lists.admb-project.org/mailman/listinfo/users</a><br>
<br>
</blockquote>
</blockquote>
<br>
</blockquote>
<br>
_______________________________________________<br>
Users mailing list<br>
<a moz-do-not-send="true"
href="mailto:Users@admb-project.org" target="_blank">Users@admb-project.org</a><br>
<a moz-do-not-send="true"
href="http://lists.admb-project.org/mailman/listinfo/users"
target="_blank">http://lists.admb-project.org/mailman/listinfo/users</a><br>
</div>
</div>
</blockquote>
</div>
<br>
</div>
</blockquote>
<br>
</body>
</html>