The following report is available in PDF format for high quality printing.

---

Second Quarter FY 2003 Report - The National Fusion Collaboratory

Edited by D.P. Schissel(1)

(1)General Atomics (schissel@fusion.gat.com)

Overview

The focus this quarter was on understanding the usersπ feedback on the FusionGrid services deployed last quarter and developing a plan on how best to smooth over some of the rough spots related to ease of joining and ease of use.

General accomplishments include:

∑             Coordinated updates on the TRANSP FusionGrid service with communication provided via the Fusion Grid Monitoring System.

∑             The second fusion code, GS2, is being tested on FusionGrid.

∑             Testing and refinements continued on the reservation based QoS system that will allow between tokamak pulse data processing to take place on FusionGrid.

∑             A demonstration of a prototype tiled display wall was done in the NSTX control room and was well received.

∑             An AG PIG is being constructed at DIII-D yielding AG nodes at the three large U.S. tokamaks.

∑             Participation in a number of conferences and workshops including the SciDAC PI meeting and the review of the DOE Science Grid Project.

General

Detailed feedback on initial FusionGrid services deployed last quarter continued to be received including new requests to join the Grid. These discussions helped to focus ideas required to make FusionGrid easier to join and the services easier to use. A section of the project website on services was created with the aim that eventually the services section will be the dominate part used to assist fusion scientists take advantage of the new capability offered by FusionGrid.

 The Collaboratory Project was represented at the SciDAC PIs meeting with a 2 page written summary of our progress for general distribution and a poster presentation to the attendees summarizing our progress to date. Immediately after the PI meeting the Collaboratory Project participated in the review of the DOE Science Grid Project.

The Collaboratory Project was involved in a number of meetings this quarter including:

∑             A presentation at the SIAM Conference on Computational Science and Engineering in a Symposium titled Plasma Physics and Computer Science: A Partnership for Advancing the Understanding of Magnetically Confined Plasmas.

∑             Several posters at the GlobusWorld conference

∑             A presentation to the ESNET Steering Committee on the Collaboratory Project and network related requirements.

∑             A presentation with demonstrations to the SciDAC Project on Numerical Computation of Wave-Plasma Interactions in Multi-dimensional Systems.

Planning began this quarter for the review of the Collaboratory Project. Initially scheduled for April in San Diego, the review will now take place May 29 & 30 at MIT. A website has been created for the review including registration, hotel information, and a draft agenda.

The Project web site continued to be updated as required.

Security/Remote Computing

Through the end of this quarter, the TRANSP code on the FusionGrid has been run 755 times taking 4,657 CPU hours while encompassing seven different projects. AT PPPL, the usage of the legacy VMS based TRANSP system has been discontinued with all TRANSP runs now done on FusionGrid. Additionally, the C-Mod group has begun to use the FusionGrid TRANSP service so that all three large U.S. tokamaks now use this new Grid service. As more individuals are using the TRANSP service some effort has focused on creating a more efficient and easier to use system. The performance of writing TRANSP output data back to MDSplus has been improved by factors of 10 to 20. Updates to the production version of TRANSP on FusionGrid are done in according to a clear set of procedures: 1) Notify users of update in progress at the Fusion Grid Monitor (FGM) web site, 2) User jobs are held in the queue pending completion of the update, 3) The successful completion is automatically detected so that held jobs are started and a completion notice is posted at FGM.

Research work continued on the ability to smoothly run between tokamak pulse data analysis on FusionGrid. This reservation based QoS system is aimed at enabling users to run fusion applications as network services with a guaranteed completion time frame. A test case, using the TRANSP code, was created for experimentation purposes. Part of this work has also included the development and installation of a Virtual Application Server along with a comparison of the performance of MDSplus and GridFTP. The ability to estimate data transfer times is one critical component of the QoS capability.

Based on the previous quarterπs design, an alpha version of the GT2 GRAM authorization package with Akenti has been implemented for testing. A paper on the new job-manager Akenti interface was presented at the CHEP03 Conference. A paper on the implementation of the GRAM authorization package was accepted for publication at the International Workshop on Grid Computing to be held in conjunction with the Middleware 2003 Conference.

The second fusion scientific code to be placed on FusionGrid is GS2, which is presently in a testing stage by several expert users. The development of this code is lead by Bill Dorland at the University of Maryland to study low-frequency turbulence in magnetized plasmas. The effort to get this code up and running involved the coordinated effort between the Collaboratory Project and Bill Dorland which included building a multi-node Linux Beowulf cluster at Maryland and the installation of MDSplus and Globus software to allow FusionGrid service. Compared to deploying TRANSP on FusionGrid, the time to deploy GS2 was substantially less since a large fraction of the groundbreaking work done on TRANSP was directly applicable to GS2.

Visualization

A prototype single PC tiled display wall was demonstrated in the NSTX control room to PPPL scientists (Fig. 1). This system included automatic alignment, Gigabit Ethernet networking, and a VNC viewer. Control room computers connected to this display wall included systems running Windows, Linux, and Macintosh operating systems. The demonstration was well received by the fusion scientists with a definite interest to install a permanent system in the NSTX control room. The automatic alignment software was newly developed this quarter and allows a single PC with multiple graphics cards to drive a display wall with excellent alignment characteristics (Fig.2). A paper titled "Color Gamut Matching for Tiled Display Walls" was accepted for presentation at IPT 2003.

Figure 1 Scientists evaluate the prototype display wall in the NSTX control

Figure 2 The left image shows the natural (unaligned) images emerging from the projectors. In contrast the right image shows the seamless image resulting from automatic projector alignment.

Tiled display work also focused on the continuing testing and development of DMX with accelerated Chromium support. This allows for accelerated graphics spread across multiple machines for rendering into a given window on a large desktop display. Additionally, some initial investigation into Synergy, a wrapping of the X2X toolkit for sharing mouse and keyboard between multiple computers, was begun this quarter.

Access Grid work focused on the creation of a PIG node at DIII-D and the refinement of the existing nodes at PPPL and C-Mod. At DIII-D, PIG equipment was purchased, delivered, and setup. Working with ESnet staff, the local network configuration is being adjusted to support the system before it can be tested. The PPPL AG node had modifications to the audio input settings and an improved audio matrix configuration. Upon completion of the DIII-D PIG, a fusion AG meeting will take place to introduce more fusion scientists to the benefits of the AG. In preparation for increased fusion science usage, a plan has been created for a Fusion Collaboratory venue and how best to integrate fusion tools into the AG. This would be a persistent space that the Fusion scientists and Collaboratory Project team members could meet and use the AG 2.0 software. Extensions to the AG capability on tiled display walls has included the continued development of remote visualization tools that have been packaged together and made available via anonymous CVS checkout. These tools are of general interest and are being downloaded by users outside of the Fusion Collaboratory Project.

 In the area of 3D visualization software, refinements were made to ELVis and research was undertaken on new techniques for visualizing uncertainties in vector fields.

Appendix A: Non-Edited Reports from Individual Institutions

A.1      M. Papka for the ANL MCS, Futures Laboratory

This quarterπs effort focused on continued development of the Access Grid and desktop uses for tiled display. Work on the Access Grid focused on continued development of remote visualization tools, ANL has packaged the tools that it has built and made them available via anonymous CVS checkout. Tools can be checked out using the following commands;

∑             setenv CVSROOT :pserver:anonymous@fl-cvs.mcs.anl.gov:/cvsroot

∑             cvs login

∑             cvs checkout flx_spu

∑             cvs checkout FLXmitter

 

from the website http://fl-cvs.mcs.anl.gov/. These tools are now being downloaded by users outside the Fusion Collaboratory. Building on the AG 2.0 architecture; we have been developing a plan for a Fusion Collaboratory venue and how to integrate Fusion tools into the AG. This would be a persistent space that the Fusion Collaboratory participants could meet and use the AG 2.0 software. ANL worked with General Atomics on the purchase of equipment needed to construct a PIG node.  Effort has begun to construct a web services interface to geeViz in order to imbed tool into other applications; the AG being one, but it maybe applicable to other Fusion efforts also.

The tiled display effort continuing the testing and development with DMX, ANL has installed and tested the latest version of DMX with accelerated Chromium support. This allows for accelerated graphics spread across multiple machines for rendering into a given window on a large desktop. ANL reported bugs discovered back to the development group. ANL tested windowed version of Chromium against SCIRun. ANL also began looking at Synergy; a wrapping of the X2X toolkit that lets you easily share a single mouse and keyboard between multiple computers. This improves on use of X2X that ANL has already been using in the shared desktop effort.

Also during this quarter the ANL held a series of teleconferenceπs with Princeton Computer Science working on long range planning and cross use of software.  ANL packaged two of its software tools for Princeton to try out on their tiled display. ANL also downloaded a package from Princeton to try on the ANL walls. The overall goal to design and build common tools that can be shared across sites.

A.2      K. Keahey for the ANL MCS, Distributed Systems Laboratory

In the last quarter we made contributions in the following areas:

 

1. Work on authorization and enforcement
1. We implemented the alpha version of the GT2 GRAM authorization package following the design prototyped in the previous year, and based on the specification developed in the previous quarter. The package was released in March. It includes documentation, installation instructions for both users and developers, as well as instructions and client and server examples for authorization system developers. The package is available at http://www-unix.globus.org/gram_authz/. It is currently scheduled to be integrated into the 3.2 release.
2. We wrote a paper describing and analyzing the implementation of the GRAM authorization package; the paper got accepted for publication at the International Workshop on Grid Computing to be held in conjunction with the Middleware 2003 conference.
3. Together with our LBNL collaborators, we prepared a presentation describing how our enforcement solution was integrated with the Akenti authorization system to the CHEP03 conference.
4. We explored technologies related to dynamic accounts and sandboxing technologies needed for more effective enforcement of the authorization techniques described above in the GT3 context.

 

2. Work on QoS-based application servers
1. We continued our work on a prototype of a network services infrastructure with a reservation-based QoS interface module enabling users to run fusion applications as network services guaranteeing a certain QoS which will enable Fusion scientists to run code between pulses. Together with our collaborators in GA and PPPL, we came up with a test case allowing the use of network services between pulses and began work in preparation to implementing it.
2. We initiated design and development allowing the use of the Virtual Application Server on the Fusion sites: the code was ported to Fusion sites, we compared the performance of MDSplus versus GridFTP, developed methods for estimating transfer times and combining them with the  resource reservations we can provide to estimate the application execution times as well as schemas, architecture and other elements necessary to carry this out.
3. We tech reported a paper describing the design of the application server, and our experiences with it (Technical Memorandum ANL/MCS-TM-262).

 

3. Assistance with Fusion Servers
1. We provided assistance on resource management aspects of Globus to  help bring up the second Fusion network service, the GS2 code, on a non-standard machine configuration.

 

4. Demonstrations, outreach and conference participation
1. We participated in the SciDAC PI meeting in Napa Valley and the preparations for it.
2. We initiated interactions with NaukaNet, enabling better network connectivity with Fusion scientists in Russia, and also started talks with this group on the potential and requirements for a Grid infrastructure to support ITER (see http://www.naukagrid.org/iter/).

 

5. General interaction
1. We provided technical support, assisted with debugging and troubleshooting of Globus installation and associated technologies. This involved providing explanations and technical discussions of Globus functionality as well as providing examples, help with troubleshooting and answering questions.
2. We also interacted with our collaborators through teleconferences, meetings, discussions, etc.

A.3      D. Schissel for the General Atomics Fusion Group

General

∑             A presentation on the National Fusion Collaboratory Project was given at the SIAM Conference on Computational Science and Engineering in a Symposium titled Plasma Physics and Computer Science: A partnership for Advancing the Understanding of Magnetically Confined Plasmas (Schissel).

∑             A poster on FusionGrid was presented at the GlobusWorld Conference (Peng).

∑             The National Fusion Collaboratory Project participated in the review of the DOE Science Grid Project (Schissel).

∑             A presentation on the Collaboratory Project and its implications for network requirements as given at the ESNET Steering Committee Meeting (Greenwald).

∑             The SciDAC project on Numerical Computation of Wave-Plasma Interactions in Multi-dimensional Systems held a workshop at GA in January. A presentation along with demonstrations was given by the NFC project on both remote computing, data management, and shared visualization. Areas of common interest were identified and will be explored in the coming months (Schissel).

∑             Preparations began for the review of the Collaboratory Project. This work included all local organizations aspects as well as preparing a detailed web page with preliminary agenda (Schissel).

∑             A poster on the Collaboratory Project was presented at the 2003 SciDAC PI meeting (Schissel).

∑             An abstract was submitted to the 2003 TTF meeting summarizing the status of the Collaboratory Project along with services that are presently available (Schissel, Greenwald).

∑             The project web site (http://www.fusiongrid.org) was maintained (Schissel).

Security/Remote

∑             A test case was created for running TRANSP between pulses. This work included creating a realistic physics analysis scenario and coordinating these results with ANL who are researching preemptive scheduling and QoS issues (Peng).

Visualization

∑             Hardware was ordered and delivered for a small AG node (PIG) at DIII-D. Node registration and installation was completed and network configuration issues including multicast are presently being addressed (Abla).

A.4      M. Thompson for the Lawrence Berkeley National Laboratory

Work was started on the new job-manager Akenti interface. We started to set up a testing environment at LBL and to sketch out the interface module. Eventually we picked up the new job-manager from Argonne and will use that for testing.

Presented a paper at CHEP03 on this work. It will be published in the Proceedings of the 2003 Conference for Computing in High Energy and Nuclear Physics.  "Fine-Grain Authorization for Job and Resource Management using Akenti and Globus", Kate Keahey, Sam Lang, Bo Liu, Mary Thompson, Von Welch, CHEP03, Mar. 2003

Continued to issue DOE Grids certificates for new Fusion Grid members and hosts: 8 people, 2 hosts.

A.5      T. Fredian for the MIT Plasma Fusion Science Center

The multigraph visualization tool used for grid-enabled TRASNP was improved this quarter. Assistance was provided to Bill Dorland of the University of Maryland with MDSplus and grid issues that enabled him to put GS2 on FusionGrid for alpha testing.

A.6      D. McCune for the Princeton Plasma Physics Laboratory

Collaboratory Computational Services:

We continued to develop and improve the TRANSP service.  As a mark of our confidence in the Collaboratory service, we have discontinued our legacy VMS-based production system, which we had maintained as a backup system for a time.  With help from Tom Fredian at MIT, we moved Cmod production from the legacy VMS system to the Collaboratory system.  Also, with the key technical contribution by Tom Fredian, the TRANSP MDSplus data write-back efficiency has been greatly improved.  We also instituted procedure for updating the TRANSP production code with minimal disruption to users, and we supported experiments for "pre-emptive scheduling" of TRANSP runs for possible future between shots analysis applications.

A bulleted list summarizing accomplishments follows:

∑             CMod joins TRANSP service.

∑             The three large US tokamak projects + PPPL Collaborations users now all use the Fusion Collaboratory TRANSP production system.

∑             Performance of MDSplus writes of TRANSP output greatly improved: factors of 10-20 improvement seen to GA MDSplus server.

∑             Improvements to TRANSP website (http://w3.pppl.gov/transp) -- new section, "TRANSP Compute Server Access".

∑             Command-line script-based TRANSP client package now available at the NTCC website: TR_CLIENT module at http://w3.pppl.gov/NTCC.

∑             Production code update procedure introduced:

o           notify users of update-in-progress at Fusion Grid Monitor

o           user jobs are held in queue pending completion of update

o           successful completion, automatically detected, releases user jobs and posts completion notice on Fusion Grid Monitor

o           GA team supported Fusion Grid Monitor improvements to make this work smoothly.

∑             Special queues have been created for experimentation with high priority "between-shots" analysis jobs.

∑             PPPL supported ANL experiments on between-shots capability.

∑             PPPL supported development of GS2 computational service, proceeding at University of Maryland.

PPPL Contributions to Development of Collaboratory Visualization Capabilities:

Work continued in three areas: Access Grid (with ANL), Collaborative Display Wall (with PCS), and ElVis (with the PPPL TRANSP team).  There was progress on all three, with highlights given in the following bulleted list:

∑             Access Grid (AG) on PPPL Display Wall (with ANL):

o           Adjusted audio input settings to improve pick-up on all microphones.

o           Received improved audio matrix configuration from ANL

o           Fixed bug in AG Beacon software.

∑             Collaborative Display Capability (with PCS):

o           Developed proposal for NSTX project, developing several configuration options.

o           Demonstrated prototype in NSTX Control Room, showing efficacy of PCS Display Wall auto-allignment software.  Demonstration also confirmed ability to attach NSTX Control Room Macintosh clients to the Display Wall system on a gigabit network.

∑             ElVis:

o           Software installation & documentation on PPPL Linux Clusters making system available to code developers.

o           Implemented intelligent self-adjusting graph tic marks (imported into Java from older fortan-based PPPL scientific graphics software).

o           Demonstrated early prototype of collaborative surface plot.

A.7      G. Wallace for the Princeton University Computer Science Department

∑             Prototyped single PC display wall alignment software. This will allow a single PC with multiple graphics cards to drive a display wall and show an aligned Windows Desktop.

 
Figure A.7-1 The left image above shows the natural (unaligned) images emerging from the projectors. In contrast the right image shows the seamless image resulting from automatic projector alignment.

 

∑             Worked with PPPL to prototyped a fusion collaborative display in the PPPL display wall room. This included automatic alignment, gigabit networking, and VNC viewer. We had computers running Windows, Linux and Mac OS all collaborating on the display-wall.

∑             Moved the prototype into the PPPL control room for a demo and feedback from fusion scientists. Feedback was positive and we have input on what to improve.

Figure A.7-2 Scientists evaluate the prototype display wall in the PPPL control room.

∑             Completed paper "Color Gamut Matching for Tiled Display Walls" accepted to IPT 2003 (Immersive Projection Technologies conference). This technique is important for seamlessly color balancing DLP and mixed arrays of projectors.

A.8      Sanderson for the University of Utah Center for Scientific Computing and Imaging

The quarter was spent investigating new techniques for visualizing error in vector fields. We have developed a method using reaction diffusion techniques which is being submitted to Visualization 2003 for publication.