|
The breveCluster at NMU uses asynchronous computation
among multiple computers for student research into evolutionary learning
and artificial life with the
breve simulation environment. Current developments
on the cluster are brought into being by Correy Kowall and Brian Krent (along with much
thanks to efforts Jon Klein for his creation, continual development, and support of the
breve software package). A list of various files used by and resulting from the project
can occasionally be found under the files folder.
What we've done...
Conducted during the Winter and Fall 2006 semesters,
Correy Kowall and Brian Krent have completed several evolution experiments of
autotrophic reproduction within the breve simulation environment on the cluster.
Initial work on this was presented at NMU's Celebration of Student Research &
Creative Works and as a department colloquium.
The results from these experiments were presented at the international Genetic
and Evolutionary Computation Conference (GECCO) held in London in July 2007.
|
What's current...
Continuation of Project Isthmus.
We are now transitioning over to an entirely new codebase
that will implement a heterotrophic environment, in which agents will be able to feed off of other agents within
the simulation. Also, under Project Synergy, detailed plans are
being drawn up (along with tons of background research being conducted)
for the next generation of breveCluster. Unfortunately, since breveCluster is currently all comprised of
borrowed equipment, 50% of the worker nodes had
to be returned to the MA/CS Lab for the Fall 2007 CS 302 class.
|
Where we're going...
Started in the summer of 2007 and still continuing onward, Project Latte, a major overhaul of breve will be ongoing for the next year,
which aims to optimize utilization of computational resources on various levels.
In the Fall 2007 semester, the first interdepartmental use of breveCluster will occur with a simulation to be
conducted by Chris Wells, who is doing an undergraduate directed study project under the guidance of Dr. Mac Strand from the Biology Department. (Preparations have already been made during the Winter 2007 semester.)
|
Status & Info
Log
| Date |
|
Notes |
| April 22, 2007 |
|
Things have been very busy recently,
which is why the log hasn't been updated. The
breveCluster was moved from NERL to the
MA/CS lab awhile back ago, and just yesterday
it was moved to its own air-conditioned room.
(Thanks to Professor David Powers for getting
the air-conditioning installed!)
Chris Wells's programming for Project Waterhole
has progressed quite nicely.
And, Correy Kowall & Brian Krent have submitted
their camera-ready copy to GECCO 2007.
|
| February 12, 2007 |
|
The adventures of ErgoMan! Correy created a hominid that uses Hebbian learning with feedback to attempt to raise the altitude of the head.
See 20070212-ErgoMan3.mpeg and 20070212-ErgoMan5.mpeg.
(Clarification note: The ErgoMan code is currently in-development and is not running on the cluster. The cluster is currently running an incarnation of the Isthmus codebase.)
|
| February 11, 2007 |
|
node02 has been reinserted into the cluster and is fully operational.
|
| February 9, 2007 |
|
A sneak-peek at work in-progress currently being done by Correy Kowall:
Body,
Rope.
(Note to the reader: This is a real-time, interactive, dynamic simulation within the breve environment and not mere predetermined animation.)
|
| January 22, 2007 |
|
Optical drive issues discovered with the returned node02,
so the entire unit will be swapped out for another one (permanently); that
returned unit will go back again for repairs, but we're not going to deal
with it ever again, since we just permanently replaced it with something
from the pool of other systems.
|
| January 16, 2007 |
|
node02 has been returned to us with a new hard disk. (Twas removed from cluster due to drive failure.)
|
| December 7, 2006 |
|
The grapher was given a slight facelift,
and standard maintenance was performed on all nodes (full system updates across the board).
|
| October 16, 2006 |
|
Real-time results grapher is now online, thanks to Alex:
brevecluster.nmu.edu/cgi-bin/grapher.cgi
|
| October 1, 2006 |
|
Some video footage of recent activity:
20060927-Isthmus-TwoArmedWithCommunication.mpeg,
20061001-Isthmus-TwoArmed.mpeg.
|
| September 25, 2006 |
|
Standard maintenance updates applied; all nodes rebooted (as required by updates).
Script change in effect on all nodes: breve cycling set to two hours (instead of 20 hours). |
| September 22, 2006 |
|
Lock file bug fixed in A.tz (problem was in the save-results method). |
| September 15, 2006 |
|
Standard maintenance updates applied; all nodes rebooted (as required by updates). Current A.tz running under breve on
node00 through node07. |
| September 14, 2006 |
|
Correy's new code is going through a trial run with breve on node01 through node07 and node00 acting as centralized data storage.
Initially there were some permissions problems on node00, but they were resolved; and everything seemed to be running
fine at the end for the day. Currently using breve version 2.5d1. If problems occur with 2.5d1, then we will download
breve 2.4 via subversion and build it ourselves, due to the issues experienced with the current 2.4 pre-compiled binaries available online. |
| September 12, 2006 |
|
Cluster rebuild progress: NFS and breve has been setup on nodes 00 through 07. Still have to do node08. |
| September 9, 2006 |
|
breveCluster "version 2.0" coming soon. All but one of the Pentium III systems were replaced with Pentium 4 systems. Still to do: SMP kernels on new
machines, NFS, etc. |
| September 6, 2006 |
|
Doing a clean install of Ubuntu 6.06 on all nodes for the Fall 2006 semester. SMP-686 kernels are currently in use on node00, node04, and node08 to squeeze what little juice we can out of Hyperthreading. |
| April 2, 2006 |
|
Added /Network/Tools to the PATH environment variable on all nodes for our standard user login. Note: File organization will
resemble that of the filesystem laid out by
GNUstep. |
| April 1, 2006 |
|
Unified breveMaster and breveSlave folders into just /breve to simplify things. Shares (/etc/exports) and BREVE_CLASS_PATH environment variable on head node updated. Share mounts (/etc/fstab) and BREVE_CLASS_PATH environment variable updated on all worker nodes to reflect this change. The local folder /breveSlaveData is now /breveData on all nodes, and C.tz was updated to reflect this change. All old, unused folders removed. Also, there is now a general purpose read-only share on the head node called /Network mounted on all worker nodes. |
| March 4, 2006 |
|
New NFS setup implemented for low-maintance sharing of breve files to the worker nodes. |
| February 2, 2006 |
|
The void which was node04 has finally been replaced with a new HP-compaq system. It's shiny. |
| January 28, 2006 |
|
Since the Windows build of breve seemed to be the offending factor on multiple levels, Linux was installed on the head node. The cluster is now 100% Linux. |
| January 25, 2006 |
|
Debugging breve 2.4 pre-alpha weirdness (still looks like breve itself is to blame; workaround devised and to be implemented tomorrow) |
| January 23, 2006 |
|
Debugging our own code (though, problem appears to be breve's fault) |
| January 22, 2006 |
|
Assembly and shakedown |
| January 20, 2006 |
|
Initial hardware gathered |
Documentation, et cetera
Disclaimer:
Any opinions, findings, and conclusions or recommendations
expressed in this material are those of the author(s) and do not
necessarily reflect the views of the National Science
Foundation.
|
|
