Notes
Outline
Center for Computational Research
All Hands Meeting - Tuesday, March 11, 2003
Russ Miller, Director
Tom Furlani, Associate Director
Agenda
Gordon E. Moore
Co-Founder of Intel
Predicted (1965/75) that transistor density would double every 12/18 months
Processing speed doubling every 18 mos.
Disk storage doubling every 12 mos.
Aggregate bandwidth doubling every 9 mos.
What is a (Beowulf) Cluster?
Industry Standard Hardware and Software
PC-Based Components (Intel or AMD)
Ethernet or Myrinet
Linux, PBS, MPI
“Commodity Off-The-Shelf” (COTS)
Introduced in 1995
Operates as a Single System
Rivals Performance of Traditional Supercomputer at a Fraction of the Price
Fastest Computers
Earth Simulator in Japan
(NEC Vector Supercomputer)
Earth Simulator
40TFlops Peak
Homogeneous, Centralized, Proprietary, Vector
Expensive!
CFD-Weather, Climate, Earthquake
640 NEC SX/6 Nodes (5120 CPUs)
Footprint = 4 tennis courts
$6M/year in power
Center for Computational Research
General Information
High-Performance Computing and High-End Visualization
47 Active Research Groups in 27 Depts
Leverage $300K NSF Grant Ž
$41M Vendor Donations
$55M Grants & Contracts (non-bldg)
Deliverables
350 Publications & Pres., Hardware, Software, Algorithms
Outreach (EOT)
Workshops, Training, Courses, Degree Programs
25 Local Companies & Institutions
Funding
NSF, NIH, NIMA, EPA, Keck, Sloan
NYS, SUNY, UB
IBM, SGI, Sun, DELL, HP (Compaq), Nortel, Myricom
Staff
Director – Dr. Russ Miller
Associate Director - Dr. Tom Furlani
Computational Scientists
Dr. Zihua Hu (Bioinformatics), Dr. Matt Jones (Physics),       Dr. Jeff Tilson (Chemistry)
Programmers
Steve Gallo (Grid Computing), Quoc Nguyen (Bioinformatics)
Visualization
Adam Koniak (Urban Simulation), Martins Innus (Scientific Visualization)
Systems Support
Cynthia Cornelius, Jon Bednasz, Sam Guercio, Tony Kew, Deborah Loke, Dori Macchioni , Jason Rappleye
Office Staff
Brenda Sauka (Financial Manager), Sally Elder (Secretary), Penelope Krebs (Receptionist)
ExCom & AdvCom
Executive Committee
Josephine Anstey, Media Study
Corky Brunskill, SENS
Philip Coppens, Chemistry
David Kofke, Chemical Engineering
Bruce Pitman, Math
Internal Advisory Committee
Bruce Holm, Senior Vice Provost
Voldemar Innus, CIO
Mark Karwan, Dean School of Engineering
Bruce McCombe, Assoc. Dean, CAS & Director, CAPEM
Space
1999
Norton 9: 2800 sq ft + 1100 sq ft Machine Room
2002
Norton 10 AGN, TDW: 600 sq ft
Norton 15 Machine Room: 1700 sq ft
Norton 118 Machine Room: 1700 sq ft
Norton 17 Office Area: 1600 sq ft
Jan., 2003 - Delayed
Norton 5 & 7: 3000 sq ft
Provide Public Access to iDesk, AGN, Videoconferencing
Provide Appropriate Staff Office Space
Provide Visitor Office Space
UPS System – Delayed from 1/02 to 3/03
Making CCR Easier to Use
Website Update
New Look/Feel by end of month
User Info/Hotpage - Demonstration
ccr-help@ccr.buffalo.edu (RT-Tracker)
Grid Software on the Horizon
Grid: flexible, secure, coordinated resource sharing among dynamic collections of individuals, institutions, and resources
Grid Portal Under Development
HPC Machines, Condor Flocks, Storage, Visualization, Instruments
Domain-Specific Areas of Portal
Computational Chemistry Portal (ECCE)
Machine Utilization
Greater than 80% on most platforms
Queuing Systems
Regularly adjusting queue structure to maximize throughput
7 day queues discontinued
3 day queue is now the longest
Throughput and utilization is significantly better
Feedback from users is welcome
Recent change under investigation: partition for parallel and sequential jobs on nash
Outreach
Collaboration with local institutions
Buffalo Neuroimaging Analysis Center – MS MRI Visualization
Children’s Hospital – Medical Imaging
BuffLink – 3D Visualization of the Buffalo Niagara Medical Campus
Peace Bridge Authority
Bioinformatics in Buffalo (UB, Roswell Park, HWI)
H.S. Bioinformatics Clusters
Numerous Local Colleges
H.S. Summer Workshop in Bioinformatics
Tours
Faculty candidates, speakers, business leaders, politicians, funding agencies
Computational Resources
SGI Origin2000 ® Origin3800 (Crosby): 1998+
64 Processors (400 MHz)
Shared memory
32 GB RAM
400 GB Fibre Channel Disk
Full Maintenance for 2 years
Primary Applications
Computational Fluid Dynamics
Quantum Chemistry (ADF/Jaguar/Q-Chem)
Computational Mechanics
Physics
Computational Resources
IBM RS/6000 SP (Stills): 1998+
78 Processors
26 GB RAM
640 GB Disk Storage
Primary Applications
Computational Fluid Dynamics  (CFX4)
Chemical Engineering (Molecular Dynamics, etc)
Quantum Chemistry (NWChem, ADF, Columbus, CHARMM, Q-Chem)
MAE (PETSC)
Off Maintenance Jan 2004
Requires Exit Strategy
Computational Resources
Sun Microsystems Cluster (Young): 1999+
48 Sun Ultra 5s (333MHz)
16 Dual Sunblades (750MHz)
30 GB RAM, Myrinet
Sun Solaris OS
Large Number of Sequential Jobs
Primary Applications:
Quantum Chemistry (ADF)
Chemical Engineering
CHARMM
Processors Showing Age
Redeploy when Opportunity Presents
Computational Resources
SGI Intel Linux Cluster (Nash): 2000
150 PIII Processors (1 GHz)
75 GB RAM, 2.5 TB Disk Storage
Myrinet2000
Primary Application Areas:
Chemical Engineering
Computational Fluid Dynamics
Off Maintenance in 1.5 years
Stability Issues due to Poor Hardware
Consider Exit Strategy
Computational Resources
New Dell Linux Cluster (Joplin): 2002
600 P4 Processors  (2.4GHz)
Myrinet2000
600 GB RAM; 40 TB Disk
22nd Fastest Computer in the World
Large Production Machine
Highly Parallel Codes
Primary Application Areas:
Chemical Engineering
Computational Mechanics
Computational Fluid Dynamics
Environmental Chemistry
Quantum Chemistry (ADF,NWChem)
Computational Resources
Sun Enterprise 6000 (Database): 1998
4 UltraSparc-II 336 MHz
2 GB RAM
38 GB disk + 2 ´ T3 Disk Arrays
Primary application areas:
Large Oracle Database Application
Data Mining – Tops Friendly Markets
Computational Resources
Apex Bioinformatics System: 2003
Sun V880 (3), 6800, 280R (2), PIIIs; Sun 3960: 7 TB Disk Storage
BioACE  (Bioinformatics Application Computing Environment)
Available Directly from CCR Website
Faculty/Staff/Students from UB, RPCI, HWI
Home to Many State-of-the-Art Bioinformatics Software Packages
DNA and Protein Sequence Analysis (on hold)
Database Search
Gene Expression Profiling
Data Visualization
Missing Small Linux Cluster (DNA & Prot. Seq. Anal.)
GCG Implementation Postponed
Workshop Tomorrow on InforMax Packages
Visualization Displays
Visualization Computing
SGI Onyx 300 – 10 Processors
Infinite Reality 4 Graphics Pipes (1GB Texture memory)
Used to drive the iDesk and the 3300W simultaneously
SGI Onyx2 – 6 Processors
Infinite Reality 2 Graphics Pipes (64MB Texture memory)
Access Grid Node (AGN)
Facilitates group to group collaboration
Budget (1)
Personnel: $1.2M/yr by UB
Discretionary Operating: £ $100K/yr by UB
Travel, training, equipment, software, maintenance, memberships, repairs, tapes, supplies, phones, installs, shipping, promotional material, networking, furniture, etc
Equipment Acquisitions: Ad Hoc
Approx. $6.5M to date (4 year period)
Comparison
Budget/year Staff
CCR $1.3M 18
NCSC not for public dissemination
OSC not for public dissemination
U Minn not for public dissemination
Budget (2)
CCR-Leveraged Grants & Contracts
CCR Grants & Contracts: $12M
Vendor Donations (in-kind): $41M
External Research Funding: $43M                        ($290M incl corporate, construction funds, foundation support, etc.)
User Support
Critical that CCR is included in routing sheet for proposals that will use facilities
Direct costs for CCR are very important in these financial times
Student Support from Operating, Workstudy, or External Grants (including REUs)
Charge-Back Scheme
Schemes for Cost-Recovery Under Consideration
Basic Rates Have Been Determined
SGI Linux: $0.4 per processor per hour
SGI Origin: $0.6 per processor per hour
IBM SP:  $1.0 per processor per hour
iDesk $69.0 per hour
What Does This Mean for Users?
 $1,000 - $22,000 per month
Status of Charge-Back Scheme Unknown
Visualization Review
External Panel Convened to Review Visualization Efforts on Campus
Site Visit on Dec 19th, 2002
Committee did not read Documents
No Debriefing with Committee
Brief Conversation with Provost in Feb.
Positive Feedback on CCR
Central Funding Should be Provided
Charge-Back Model is Not an Option
Formation of Provost Advisory Panel for CCR
Consolidation & Upgrade Plan
Retain
Dell Linux P4 Cluster
Acquire
1TF SMP System (Regatta, Alpha, Itanium)
0.5TF General Purpose PC farm
Storage Area Network (SAN)
Evaluate
Sun6000 Database System
Retire/Redeploy
IBM SP
Sun Cluster
SGI Linux Cluster
Derailed with Cancellation of $6.5M HP EV7 & Intel Systems (PC Farm & BioCCR System)
Future Plans
Finances Restrict Planning
Needs Assessment
Stable and Reasonable Operating Budget
Level Similar to NCSC, U Minn, OSC
Will Allow for Strategic Planning
Constant Refresh of Equipment
Travel, Training, Workstations, Supplies
Software, Tapes, Cables
???Administrative Discussion???
Break
Relax, Converse, Food/Coffee
10:30 Overview of Projects at CCR - Furlani
11:15 Open Discussion/Networking
12:00 Lunch (on your own)
1:00 Intro to TDW and 3300W – Innus
1:30 Intro to AGN – Tilson
2:00 Parallel Computing at CCR – Jones/Tilson
3:00 Overview of Joplin - Cornelius/Green
4:00 Intro to Bioinformatics - Hu