Focus Effort Title: HLA Integration for HPC Applications applied to CMS.

PI Name: David Bernholdt
PI EMail Address: bernholdt@npac.syr.edu
PI Telephone: 315 443 3857        PI Fax: 315 443 1973

Description:

This proposal is in response to two related goals. Firstly the need
for message passing based coarse grain parallelism for CMS and
secondly integration of classic HPC applications into FMS
simulations. The latter is essentially equivalent to HLA compliant
versions of these applications. Both goals can be elegantly addressed
by using a computing environment built around distributed (or meta)
computing built on top of HLA.

From the parallel architectures point of view, this project will
develop distributed memory version of CMS which can be then easily
ported to distributed memory MPPs such as SP-2.  From the
metacomputing management point of view, this task develops and tests
tools for runtime management of distributed resources, initially
prototyped in a controlled commodity cluster envioronment, and in the
next stage to be expanded to a multi-MSRC metacomputing
platform. Finally, from the parallel hardware availability point of
view, this task will effectively bring NPAC commodity cluster,
including a mix of Linux and NT PCs, as one of the HPC resources
available for our Metacomputing CMS experiments. 

Accomplishments:

This task is being pursed along the following two complementary prongs:

1. Distributed memory version of CMS. We started with porting CMS to
   Linux. Since GNU C++ compiler that undelies the Linux environment
   is also available on all other UNIX boxes, such port would facilitate
   experiments with heterogeneous workstation clusters. Unfortunately,
   the current CMS code seems to be highly sensitive to numerous
   idiosyncracies of IRIX and SGI C++ compiler. GNU compiler, when
   tried on CMS, produces a long list of errors, some of which
   are quite difficult to track. We are proceeding with this task
   and we are interacting with Ft. Belvoir engieers who are also
   trying to port CMS to Linux.

2. JWORB/OWRTI based Metacomputing FMS Support. Here we started with
   a recent visit to CEWES by the end of June where we met with
   Jay Cliburn and Mitch Baker to discuss security and other system
   issues related to a sustainable multi-MSRC metacomputing environment.
   We concluded that the most practical approach will be to setup
   a kerberized JWORB servers in all centers, initially in the user mode
   (and hence to be manually restarted daily via secure id logins to
   individual labs) and later on in the system based 24x7 mode after
   WebHLA is approved by the Program Office as a viable Metacomputing
   platform for multi-MSRC job decomposition, scheduling and management.
   We also discussed job schedulers such as GRD, NQE, PBS or Miser
   used by various MSRCs and DCs and we concluded that a uniform
   meta-level scheduler packaged as CORBA service in JWORB would offer
   a natural solution to the current diversity of local scheduling
   strategies. We initiated a dialog with German company Genias that
   markets Codine/GRD and already developed some CORBA API that could
   be relevant for our purposes. 

Problems: 

none