Thomas Ndousse

Slide1: 

DOE Workshop On Ultra high-Speed Transport Protocols and Network Provisioning April 10-11 Argonne National Laboratory Terabits Networking Randamp;D for Petascale Sciences Thomas D. Ndousse, PhD Program Manager, Network Research Office of Science, US Department of Energy

Slide2: 

US Department of Energy Office of Science Advanced Scientific Computing Research Division Mathematical, Informational, Computational Sciences Randamp;D Computer Science Applied Mathematics Collaboratories Network Research Research Facilities Energy Sciences Network Supercomputing Centers ACRFs

Slide3: 

Program Mission Research to develop and deploy advanced network capabilities that will to address the unique network requirements of the Office of Science Accelerate the adoption of emerging network technologies into production networks through testing and advanced deployment Network Research Program Program Elements Basic Randamp;D Advanced Deployment and Testing Integrated Network, Middleware, and Application Activities

Slide4: 

FY-03 Network Research Activities

Slide5: 

MICS Programs • Nanoscience Materials • Chemistry • Combustion • Accelerator • High energy Physics Nuclear physics • Fusion • Climate • Astrophysics Biology • Applied Mathematics Computer Science • Network Environment Scientific Applications Genomes to Life …Applications BES, BER, FES, HEP, NP • Integrated Software Infrastructure Centers (Mathematicians, computer scientists, application scientists, and software engineers) High Performance Computing and Network Facilities for Science Research to enable… Grid enabling research Nanoscience Next Generation Architecture

Slide6: 

Research Space Randamp;D Issues not addressed by industry and other federal Randamp;D programs No economic payoff for industry to develop specialized network technologies Mission-Oriented Agency Unique network requirements to support the agency’s science mission Research focus and agenda driven by big impact science applications Integrated Network Research Integrated Randamp;D approach which include network research, middleware research, application development Randamp;D results are directly implemented on a high-performance production network Why DOE does Network Randamp;D

Slide7: 

~0.6-2.5 Gb/s Tier 1 Online System Offline Farm, CERN Computer Ctr ~25 TIPS FNAL Center IN2P3 Center INFN Center RAL Center Institute Institute Institute Institute ~0.25TIPS Workstations ~ 100 MBytes/sec 100 - 1000 Mbps Network Requirements: Ultra high-speed data transfer protocols On-demand bandwidth Physics data cache ~ PByte/sec ~2.5 Gbits/sec ~0.6-2.5 Gbps Tier 3 Experiment CERN/Outside Resource Ratio ~1:2 Tier0/( Tier1)/( Tier2) ~1:1:1 Tier 4 Tier 2 Tier 0/1 High Energy and Nuclear Physic – I

Slide8: 

Pbytes Storage Pbytes/Sec PB/Sec 80 TFlops High-Performance vis Remote vis TB/sec TB/sec GB/sec GB/sec GB/sec GB/sec High Energy and Nuclear Physic – II

Slide9: 

HENP Lambda Grids: Fibers for Physics Summary: Providing Switching of 10 Gbps wavelengths within 2-3 years; and Terabit Switching within 5-7 years would enable 'Petascale Grids with Terabyte transactions', as required to fully realize the discovery potential of major HENP programs, as well as other data-intensive fields. Problem: Extract 'Small' Data Subsets of 1 to 100 Terabytes from 1 to 1000 Petabyte Data Stores Fact: Survivability of the HENP Global Grid System, with hundreds of such transactions per day (circa 2007-2010) requires that each transaction be completed in a relatively short time

Slide10: 

Remote Visualization Challenges Typical climate model in US (2005) We need faster ~100x faster network connectivity For 'smaller' mesh sizes (6 million mesh points), with 1,000 steps, and 10 variables, we need ~10Gbit/sec transfer. (4mins) For 'larger' mesh sizes generating 4TB of data/run (GTC), we need 533Gbit/sec transfers. Plasma Physics Projects

Slide11: 

Large-scale Collaboration Challenges Cyber security for large-scale national and international collaboration Automated resource discovery and scheduling for large-scale collaboration Network performance prediction and diagnosis 5 Special Purpose Facilities 18 Scientific User Facilities 1,400 Research Projects 200 Institutions 5 High Energy and Nuclear Physics ATLAS BaBar D0 National Virtual Observatory SATR

Slide12: 

Capacity Units 1995 2000 2005 2010 Gflops Gbytes Tflops Pbytes Pflops Ebytes 10e06 10e09 10e15 10e12 Mbps Backbone Gbps Backbone Tbps Backbone Terascale Computing Gigascale Computing Petascale Computing Computing and Communication Gap

Slide13: 

Commercial IP Networks QWEST ATandamp;T MCI/UUNET SPRINT Internet Commercial IP Network 7x24 operations No QoS – Best Effort Networking Space

Slide14: 

ESnet BACKBONE Early 2003 SINGAPORE FRANCE CERN QWEST ATM AUSTRALIA 35 Major Sites 1.3 TB/Month Large Data transfer andamp; other advanced capabilities POS, SecureNet, OC-192 Core Shared link to EU

Slide15: 

Components of End-to-End Networking OS NICs System Bus Transport TCP/UDP ? OS NICs System Bus Transport TCP/UDP ? Terabits Core Terabits/sec Apps Host bottleneck Workshop Focus Workshop Focus Middleware Science Applications Middleware Science Applications Mbps/sec Transport

Slide16: 

The workshop Focus on two Major Issues Ultra high-speed transport Protocols Dynamic Provisioning of Core Networks Objective A assessment of transport protocols and network provisioning for for large-scale science Develop a near term (3 years) and long-term Randamp;D ( 5 years) framework for transport technologies that will support DOE’s emerging networking requirements Scope Focus on ultra high-speed network transport technologies for large-scale science applications This workshop is not intended to discuss the replacement of the Internet protocols Workshop Outcome A workshop report Special Issue on Communications

Slide17: 

Terabits Networks Transport Protocols Develop the framework for a versatile and stable transport protocol that can deliver and sustain Terabits/sec throughput to high-end scientific applications Outline the essential features including capabilities, functionalities, and interoperability issues of the Terabits/sec transport protocols What type of modifications are required in the core network to support Terabits/sec transport protocols? Can existing transport protocols such as TCP/variants be enhanced to satisfy the requirements Terabits networks.

Slide18: 

Why radical Provisioning strategies are Critical Scaling Issues: The current network provisioning models used in research networks may be broken and may not scale to terabits network Load Balancing: High capacity networks such as 40 Gbps (OC-768) and beyond may consist of multiple wavelengths that will introduce load balancing problems to transport protocols On-Demand Bandwidth: Transactions on ultra high-speed networks will be short-lived in contrast to 7x24 operations New Networking Business Models: Market conditions in the telecommunications sector could be exploited to develop dynamic networks that are responsive to large-scale science

Slide19: 

Provisioning: On-demand Bandwidth 50 TB from SLAC to FNL on 03/9/05 Remote vis 09/5/05 for 3 hours 500 TB from SLAC to FNL on 03/9/05

Slide20: 

Decoupling the transport protocol from the Core Networks

Slide21: 

Provisioning: On-demand Lambda Networks Provisioning: On-demand Lambda Networks On-demand lambda networks

Slide22: 

Terabits Network Testbeds Testing of terabits network transport protocols and other ultra high-speed network components Testing of dynamic provisioning concepts for terabits backbone networks Testing of ultra high-speed middleware services such FTP, IDS, etc Testing of Terascale science applications on Terabits networks

Slide23: 

Qandamp;A