Jack Dongarra University Distinguished Professor Innovative Computing Laboratory Electrical Engineering and Computer Science Department The University of Tennessee dongarra@cs.utk.edu October 1, 2008 Silvio Migliori Funzione Centrale Informatica Lungotevere Thaon di Revel, n. 76 00196 Roma Dear Silvio, This is a note on the current status of the Central Supercomputer at ENEA CRESCO as of October 1 st , 2008 General Specification The hardware specification for the high performance computer system has three basic section. Section 1 (Grande Memoria) • 42 Node SMP IBM x3850 ‐ M2 • with 4 socket Xeon Quad ‐ Core Tigerton E7330 • 32/64 GByte RAM per node • (2.4GHz/ 1066MHz/6MB L2) • 9.6Gflop/s per core • A total of 672 core Intel Tigerton Section 2 (Alto Parallelismo) • 256 node blades IBM HS21 • with 2 socket Xeon Quad ‐ Core Clovertown E5345 • 16 GB RAM per node • (2.33GHz/1333MHz/8MB L2) • 9.32Gflop/s per core EECS Department 1122 Volunteer Blvd., Suite 413 Knoxville, TN 37996-3450
• A total of 2048 core Intel Clovertown Section 3 (Special) • 4 node blades IBM QS21 with 2 socket Cell BE Processors 3.2 Ghz each. • 6 node IBM x3755, 8 Core AMD 8222 with FPGA VIRTEX5 • 4 node IBM x 3755, 8 core AMD 8222 with NVIDIA Quadro FX 4500 X2 • 4 Nodi windows 8 core 16 Byte RAM • Networking o Infiniband 4X to interconnect all nodes of the systems. • Software o LINUX and Windows support o Middleware � OpenAFS � LSF � GPFS • Benchmarks o Standard benchmarks such as HPL and HPCC Benchmarks runs • File system o 120 TB raw with 96 TB RAID6 o 3 GB/s transfer o File system GPFS at 2GB/s
General Comments on the New System CRESCO ( C entro Computazionale di R ic E rca sui S istemi Co mplessi, Computational Research Center for Complex Systems) is an ENEA Project, co-funded by the Italian Ministry of Education, University and Research (MIUR). The CRESCO project, located in the Portici ENEA Center, aims to develop a High Performance Computing infrastructure which will be mainly devoted to the study of Complex Systems. There are three major components: • The implementation of innovative solutions in topic of computing system architectures and GRID computing for the activities of advanced R&D in ENEA, requiring the use of high performance computational resources. • The study of biologic objects from a "systemic" point of view and the study of natural systems (animal and social communities) according to paradigm of complex systems; • The study of technological complex systems and their mutual interactions and the creation of suitable tools for the modeling, simulating and control processes. Funding for the facility is at the level of about 9 Million Euros and comes from the Ministero dell’Università e della Ricerca. The Project has participation from 18 partners between University, Agencies of Research, and Industry. The supercomputer facility has a peak performance of about 24 TeraFlop/s, and today is placed at position 180 on the Top500 list of Supercomputers. The inauguration of the Central Supercomputer at ENEA CRESCO took place on May 6 th , 2008. The systems and configurations specified are an important addition to computing in ENEA, Italy, and the EC in general. The overall procedure for implementing and installation of the system was well executed and the facility has been installed and currently undergoing final testing before acceptance. It is clear that the careful planning in both the hardware and software components of the computing facility as well as the necessary human infrastructure to support the systems has paid off with the smooth installation and startup of the facility. In addition, ENEA has put in place the physical infrastructure needed to house the system, and has selected a computing system that can perform well on a broad range of applicability in science and engineering. The CRESCO center has quickly come up and is in operation for a small set of important users. These users are engaged in state of the art research. Three of the researchers presented their results at a workshop at ENEA Portici on September 30, 2008. • Massimo Celino talked about “New Materials and Complex Molecules: From The Design To Your Computer”. He looked at a study of Mg ‐ MgH2 interface using Car ‐ Parrinello molecular dynamics code on ENEA ‐ GRID. • F.R. Picchia, talked about “High Performance Computing for Developing Advanced Energy Technologies” • Adriana Carillo, discussed “High Resolution Mediterranean Climate Simulations: Present and Future Scenarios.” Each talked presented research obtained from the used of the CRESCO system. In addition there were a number of presentations made by various projects in the ENEA CRESCO effort. The Projects have as its objectives the achievement at the ENEA Research Center of Portici,
an important calculation multidisciplinary study of complex systems and biological technology, combined with facilities for processing, storing and displaying 3D data Centers located at ENEA Brindisi and Trisaia. In terms of applications and scientific content, the project focuses on the following lines of areas: • Activity 1 o The deployment of innovative solutions in the field of architecture computing systems and GRID computing for R & D edge at ENEA requiring the use of computational resources extremely important. o On the infrastructure project envisages the creation of a platform for calculation of absolute excellence capable of providing a power of the TeraFLOPS, with a new class of innovative features GRID. Talks were presented by: • S. Pierattini: Tools and Solutions for 3D Graphics Developed in CRESCO • P. Palazzari: Efficient use of FPGA on CRESCO, Porting the MIT Global Circulation Model on the CellBE Processor. • S. Raia: Innovative Tools for ENEA GRID • Activity 2 o The study of biological objects from a "systemic" (in silico cell) and the study of natural systems (community and social animals) as the paradigm of complex systems. A talk were presented by: • G. Aprea: Estimating Kinetic Parameters of Metabolic Networks • Activity 3 o The study of complex technological systems and their mutual interactions, and implementation of appropriate tools for modeling, simulation and control. Talks were presented by: o M. Minichino: Study and Analysis of Complex Networks and their Implementations are ENEA ‐ GRID o S. Di Blasi: Network Simulator 2 ‐ Installation is ENEA ‐ GRID and Main Applications o S. De Porcellinis: Federating Environment Testbed for Critical Infrastructure Simulation o M. Pollino: ISIS – Territorial Information for the Protection Of Infrastructure and Elsewhere
Recomm mendations: Ensuring g usage by ap pplication is the next step p and will be e the most c ritical issue facing the successfu ul usage of th hese system s. The appro oach being t taken by EN NEA is techni ically sound d and the ENEA ha as assembled d a strong te chnical team m that is capa able of execu uting the pro oject. Although h there are ri isks associat ed with any project of th his scope and d complexity y, such as pi icking a credible v vendor, IBM M in this case e. ENEA ha as put into pl lace a manag gement appr oach based o on sound project m management principles. ENEA m must play a pr roactive role e in helping the scientific c and engine eering comm munities prep pare for the high p performance e computing g system. Tw wo actions are e suggested: : – It t is recomme ended that EN NEA engage e each princ ipal investig gator of the H HPC system, , as well as s any others selected to u use the ENE EA computin ng system, to o determine h how their pr rojects w will use the re esources pro ovided by EN NEA and wh hat level of te echnical and d other suppo ort will b e required to o make full u use of the EN NEA system ms. Were imp portant scien ntific and te echnological l areas that a are not repres sented, the E ENEA mana gement shou uld be explic citly em mpowered to o engage the e leading res earchers in t these areas, encouraging g them to app ply for ENEA compu E uting resourc ces through the appropri iate mechani isms. Course es and trainin ng sh hould be pro ovided to hel lp users adap pt their appli ications for t the new syst tems. – A As an integra al part of any y advanced c cyberinfrastru ucture, CRE ESCO should d plan to lev erage, in ntegrate, and d extend the existing use r service and d support inf frastructure t to meet the n needs of th he existing u user commun nity. Consult ting and train ning are the primary me chanisms fo or ad ddressing us ser issues tha at are not cov vered by doc cumentation n or that are r related to sy stems pr roblems. – It t is also reco ommends tha at the ENEA A host an ann nual worksho op, beginnin g towards th he end of th he year or ea arly next yea ar. This work kshop should d be focused d on identifyi ing and furth hering th he developm ment of the co omputing tec chnologies n needed to sup pport HPC c computing as s well as th he science an nd engineeri ng applicatio ons that are taking advan ntage of or c could take ad dvantage o f the comput ting resourc es. – C CRESCO sho ould offer cla asses and sem minars on hi igh performa ance scientif fic computin ng and h ands-on tuto orials on port ting and opt imizing code e for the EN NEA superco omputer. Sincerely y, Jack Dong garra University y Distinguish hed Professo or
Recommend
More recommend
