Year End Report on the NSF OptIPuter ITR Project NSF ANIR Division Arlington, VA December 12, 2002 Dr. Larry Smarr Director, California Institute for Telecommunications and Information Technologies Harry E. Gruber Professor, Dept. of Computer Science and Engineering Jacobs School of Engineering, UCSD
43
Embed
Year End Report on the NSF OptIPuter ITR Project NSF ANIR Division Arlington, VA December 12, 2002 Dr. Larry Smarr Director, California Institute for Telecommunications.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Year End Report on theNSF OptIPuter ITR Project
NSF ANIR DivisionArlington, VA
December 12, 2002
Dr. Larry Smarr
Director, California Institute for Telecommunications and Information Technologies
Harry E. Gruber Professor,
Dept. of Computer Science and Engineering
Jacobs School of Engineering, UCSD
The Move to Data-Intensive Science & Engineering-e-Science Community Resources
ATLAS
Sloan Digital Sky Survey
LHC
ALMA
CONTROL
PLANE
Clusters
DynamicallyAllocatedLightpaths
Switch Fabrics
PhysicalMonitoring
Apps Middleware
A LambdaGrid Will Be the Backbone for an e-Science Network
Source: Joe Mambretti, NU
Just Like in Computing --Different FLOPS for Different Folks
DSL GigE LAN
C
A
B
A -> Need Full Internet Routing
B -> Need VPN Services On/And Full Internet Routing
C -> Need Very Fat Pipes, Limited Multiple Virtual Organizations
Source: Cees Delaat
Number of users
Bandwidth consumed
Response—The OptIPuter Project
OptIPuter NSF Proposal Partnered with National Experts and Infrastructure
Vancouver
Seattle
Portland
San Francisco
Los Angeles
San Diego(SDSC)
NCSA
SURFnet CERNCA*net4
AsiaPacific
AsiaPacific
AMPATH
PSC
Atlanta
CA*net4
Source: Tom DeFanti and Maxine Brown, UIC
NYC
TeraGrid DTFnet
CENIC
Pacific LightRail
Chicago
UICNU
USC
UCSD, SDSUUCI
The OptIPuter Is an Experimental Network Project
• Construction of a DWDM-Based WAN Testbed– Capability To Configure Lambdas In Real-time– Scalable Linux Clusters– Large Data Resources
• Distributed Visualization and Collaboration Applications for a Bandwidth-Rich Environment
• Leading Edge Applications Drivers, Each Requiring On-Line Visualization Of Terabytes Of Data– Neurosciences Data Analysis– Earth Sciences Data Analysis
• Optical Component Research– Shaya Fainman, UCSD– Sadik Esener, UCSD– Alan Willner, USC– Frank Shi, UCI– Joe Ford, UCSD
• Optical Networking Systems– George Papen, UCSD– Joe Mambretti, Northwestern University– Steve Wallach, Chiaro Networks, Ltd.– George Clapp, Telcordia/SAIC– Tom West, CENIC
• Data and Storage– Yannis Papakonstantinou, UCSD– Paul Siegel, UCSD
• Clusters, Grid, and Computing– Alan Benner, IBM eServer Group, Systems Architecture and Performance department – Fran Berman, SDSC director– Ian Foster, Argonne National Laboratory
• Generalists– Franz Birkner, FXB Ventures and San Diego Telecom Council– Forest Baskett, Venture Partner with New Enterprise Associates– Mohan Trivedi, UCSD
First Meeting February 6-7, 2003
The First OptIPuter Workshopon Optical Switch Products
• Hosted by Calit2 @ UCSD– October 25, 2002– Organized by Maxine Brown (UIC) and Greg Hidley (UCSD)– Full Day Open Presentations by Vendors and OptIPuter Team
• Examined Variety of Technology Offerings:– OEOEO
– TeraBurst Networks
– OEO– Chiaro Networks
– OOO– Glimmerglass– Calient– IMMI
Coherence
DRAM - 4 GB - HIGHLY INTERLEAVEDMULTI-LAMBDAOptical Network
VLIW/RISC CORE40 GFLOPS
10 GHz
240 GB/s24 Bytes wide
240 GB/s24 Bytes wide
VLIW/RISC CORE 40 GFLOPS 10 GHz
...
2nd LEVEL CACHE8 MB
2nd LEVEL CACHE 8 MB
CROSS BAR
DRAM – 16 GB64/256 MB - HIGHLY INTERLEAVED
640GB/s
OptIPuter Inspiration--Node of a 2009 PetaFLOPS Supercomputer
Updated From Steve Wallach, Supercomputing 2000 Keynote
5 Terabits/s
Global Architecture of a 2009 COTS PetaFLOPS System
I/O
ALL-OPTICAL SWITCH
Multi-DieMulti-Processor
1
23
64
63
49
48
4 516
17
18
32
3347 46
128 Die/Box4 CPU/Die
10 meters= 50 nanosec Delay
...
...
...
...
LAN/WAN
Source: Steve Wallach, Supercomputing 2000 Keynote
Supercomputer Design 2010Semi-Conductor & System Trends
• Billions Of Transistors– Multiple Processors On A Die– On Board Cache And Dram Memory (PIM)– Latency To Memory Scales With Clock (Same Die)
• System Characteristics– Speed Of Light Becomes Limiting Factor In The Latency
For Large Systems – “c” Does Not Scale With Lithography
– Systems Become GRID Enabled
Source: Steve Wallach, Chiaro Networks
WAN & LAN Bandwidth Are Converging
0.1
1
10
100
1000
1998 2001 2004 2007
Year - General Availability
Ban
dwid
th -
Gbi
ts/s
ec
WANLAN
Source: Steve Wallach, Chiaro Networks
Convergence of Networking Fabrics
• Today's Computer Room– Router For External Communications (WAN)– Ethernet Switch For Internal Networking (LAN)– Fibre Channel For Internal Networked Storage (SAN)
• Tomorrow's Grid Room– A Unified Architecture Of LAN/WAN/SAN Switching– More Cost Effective
– One Network Element vs. Many
– One Sphere of Scalability– ALL Resources are GRID Enabled
– Layer 3 Switching and Addressing Throughout
Source: Steve Wallach, Chiaro Networks
The OptIPuter Philosophy
“A global economy designed to waste transistors, power, and silicon area
-and conserve bandwidth above all- is breaking apart and reorganizing itself
to waste bandwidth and conserve power, silicon area, and transistors."
George Gilder Telecosm (2000)
Bandwidth is getting cheaper faster than storage.Storage is getting cheaper faster than computing.
Exponentials are crossing.
From SuperComputers to SuperNetworks--Changing the Grid Design Point
• The TeraGrid is Optimized for Computing– 1024 IA-64 Nodes Linux Cluster– Assume 1 GigE per Node = 1 Terabit/s I/O– Grid Optical Connection 4x10Gig Lambdas = 40 Gigabit/s– Optical Connections are Only 4% Bisection Bandwidth
• The OptIPuter is Optimized for Bandwidth– 32 IA-64 Node Linux Cluster– Assume 1 GigE per Processor = 32 gigabit/s I/O– Grid Optical Connection 4x10GigE = 40 Gigabit/s– Optical Connections are Over 100% Bisection Bandwidth
Data Intensive Scientific Applications Require Experimental Optical Networks
• Large Data Challenges in Neuro and Earth Sciences– Each Data Object is 3D and Gigabytes– Data are Generated and Stored in Distributed Archives– Research is Carried Out on Federated Repository
• Requirements– Computing Requirements PC Clusters– Communications Dedicated Lambdas Over Fiber– Data Large Peer-to-Peer Lambda Attached Storage – Visualization Collaborative Volume Algorithms
• Response– OptIPuter Research Project
The Biomedical Informatics Research Network a Multi-Scale Brain Imaging Federated Repository
BIRN Test-bedsBIRN Test-beds::Multiscale Mouse Models of Disease, Human Brain Morphometrics, and Multiscale Mouse Models of Disease, Human Brain Morphometrics, and
FIRST BIRN (FIRST BIRN (10 site project for fMRI’s of Schizophrenics)10 site project for fMRI’s of Schizophrenics)
NIH Plans to Expand to Other Organs
and Many Laboratories
Microscopy Imaging of Neural TissueMarketta Bobik Francisco Capani & Eric Bushong
Confocal image of a sagittal section through rat cortex triple labeled for
glial fibrillary acidic protein (blue), neurofilaments (green) and actin (red)
Projection of a series of optical sections through a Purkinje neuron
revealing both the overall morphology (red) and the dendritic spines (green)
http://ncmir.ucsd.edu/gallery.html
Interactive Visual Analysis of Large Datasets --East Pacific Rise Seafloor Topography
• Near-term: Build Software To Support Advancement Of Applications With Traditional Models– High Speed IP Protocol Variations (RBUDP, SABUL, …)– Switch Control Software For DWDM Management And Dynamic Setup– Distributed Configuration Management For OptIPuter Systems
• Long-Term Goals To Develop: – System Model Which Supports Grid, Single System, And Multi-System Views– Architectures Which Can:
– Harness High Speed DWDM– Present To The Applications And Protocols
– New Communication Abstractions Which Make Lambda-Based Communication Easily Usable
– New Communication & Data Services Which Exploit The Underlying Communication Abstractions
– Underlying Data Movement & Management Protocols Supporting These Services
– “Killer App” Drivers And Demonstrations Which Leverage This Capability Into The Wireless Internet
Source: Andrew Chien, UCSD
OptIPuter System Opportunities
• What’s The Right View Of The System?• Grid View
– Federation Of Systems – Autonomously Managed, Separate Security, No Implied Trust Relationships, No Transitive Trust
– High Overhead – Administrative And Performance– Web Services And Grid Services View
• Single System View– More Static Federation Of Systems– A Single Trusted Administrative Control, Implied Trust Relationships,
Transitive Trust Relationships– But This Is Not Quite A Closed System Box
– High Performance– Securing A Basic System And Its Capabilities
– Communication, Data, Operating System Coordination Issues
• Multi-System View– Can We Create Single System Views Out Of Grid System Views?– Delivering The Performance; Boundaries On Trust
Source: Andrew Chien, UCSD
OptIPuter Communication Challenges
• Terminating A Terabit Link In An Application System– --> Not A Router
• Parallel Termination With Commodity Components– N 10GigE Links -> N Clustered Machines (Low Cost)– Community-Based Communication
• What Are:– Efficient Protocols to Move Data in Local, Metropolitan, Wide Area?
– High Bandwidth, Low Startup– Dedicated Channels, Shared Endpoints
– Good Parallel Abstractions For Communication?– Coordinate Management And Use Of Endpoints And Channels– Convenient For Application, Storage System
– Secure Models For “Single System View”– Enabled By “Lambda” Private Channels– Exploit Flexible Dispersion Of Data And Computation
Source: Andrew Chien, UCSD
OptIPuter Storage Challenges
• DWDM Enables Uniform Performance View Of Storage– How To Exploit Capability? – Other Challenges Remain: Security, Coherence, Parallelism– “Storage Is a Network Device”
• Grid View: High-Level Storage Federation– GridFTP (Distributed File Sharing)– NAS – File System Protocols– Access-control and Security in Protocol– Performance?
• Single-System View: Low-Level Storage Federation– Secure Single System View– SAN – Block Level Disk and Controller Protocols– High Performance– Security? Access Control?
Two Visits Between UIC’s Jason Leigh and UCSD’s NCMIR
• NCMIR Provided EVL: – EVL Will Prepare Data For Visualization On Tiled Display Systems.– With Large Mosaics And Large Format Tomography Data
• EVL Has Provided NCMIR with:– ImmersaView (Passive Stereo Wall Software) to Use In SOM
Conference Room Passive Stereo Projection System – System Has Been Installed And Is Working
– SOM Investigating Use Of Quanta Memory To Memory (UDP Based) Block Data Transfer Protocol For A Number Of Applications
• EVL and NCMIR Are:– Looking Into Adopting Concepts/Code From Utah's Transfer Function
GUIs For Displaying Voxel Visualization On Display Walls– Have Made Plans To Collaborate On The Development Of The
Physical Design Of The SOM IBM 9M Pixel Active 3D Display
Similar Results for SIO
OptIPuter is Exploring Quanta as a High Performance Middleware
• Quanta is a high performance networking toolkit / API.• Reliable Blast UDP:
– Assumes you are running over an over-provisioned or dedicated network.
– Excellent for photonic networks, don’t try this on commodity Internet.– It is FAST!– It is very predictable.– We give you a prediction equation to predict performance. This is
useful for the application.– It is most suited for transfering very large payloads.– At higher data rates processor is 100% loaded so dual processors
are needed for your application to move data and do useful work at the same time.
Source: Jason Leigh, UIC
Reliable Blast UDP (RBUDP)
• At IGrid 2002 all applications which were able to make the most effective use of the 10G link from Chicago to Amsterdam used UDP
• RBUDP[1], SABUL[2] and Tsunami[3] are all similar protocols that use UDP for bulk data transfer- all of which are based on NETBLT- RFC969
• RBUDP has fewer memory copies & a prediction function to let applications know what kind of performance to expect.– [1] J. Leigh, O. Yu, D. Schonfeld, R. Ansari, et al., Adaptive
Networking for Tele-Immersion, Proc. Immersive Projection Technology/Eurographics Virtual Environments Workshop (IPT/EGVE), May 16-18, Stuttgart, Germany, 2001.
– [2] Sivakumar Harinath, Data Management Support for Distributed Data Mining of Large Datasets over High Speed Wide Area Networks, PhD thesis, University of Illinois at Chicago, 2002.
• First University Owned and Operated State Fiber Infrastructure – Indiana University Bloomington – Indiana University Purdue University Indianapolis (IUPUI)– Purdue University’s West Lafayette Campus– Internet2 Network Operations Center
• Funded with a $5.3 Million State Appropriation 1999• I-Light Network Launched on December 11, 2001• Currently 1 and 10 GigE Lambdas
I-Light Campus Commodity Internet Usage Is Approaching 1 Gbps
PurdueIU IHETS/ITN
Outbound
Inbound
A Representation Of The Growth In
Theoretical Capacity of The Connection Between IUB And
IUPUI
Assumes All Fibers Lit Using Advanced DWDM Running Multiple 10Gbps Lambda On Each Fiber
Total Bandwidth as of January 2002
Owning Fiber Allows for Large Multi-Year Bandwidth Capacity Growth
Source: Indiana University
• Fifteen Countries/Locations Proposing 28 Demonstrations: Canada, CERN, France, Germany, Greece, Italy, Japan, The Netherlands, Singapore, Spain, Sweden, Taiwan, United Kingdom, United States
• Applications Demonstrated: Art, Bioinformatics, Chemistry, Cosmology, Cultural Heritage, Education, High-Definition Media Streaming, Manufacturing, Medicine, Neuroscience, Physics, Tele-science