Network Environment for Science Enterprise Preliminary Considerations

1. Network EnvironmentforScience EnterprisePreliminary Considerations Walt Polansky AG Meeting January, 22, 2003

2. Vu-graph Package Goals • Establish themes for Office of Science network environment “story” • Compare and contrast ESnet with similar networks. • Document present utilization of ESnet. • Identify major bottlenecks in networks. • Sketch a 5-yr. vision for networking environment.

3. Global Infrastructure for Science Genomes To Life Supercomputing andLarge-Scale Storage Advanced Chemistry High Speed Networks Computing and Storageof Scientific Groups High Energy Physics Advanced Engine Design Spallation Neutron Source Macromolecular Crystallography Advanced Photon Source

4. Office of ScienceDistributed Resources, Distributed Expertise Supercomputing andLarge-Scale Storage Advanced Chemistry High Energy Physics CMS Testbed Computing and storage for scientific groups Advanced Photon Source Advanced Engine Design Macromolecular Crystallography

5. The context- Where we are today 1995 – SC’95 I-Way demonstration • Experimental high performance network • Linked supercomputers (combined peak computing power approaching a teraflop) and visualization • A variety of applications using computing tools such as MPI for message passing, Nexus, Fortran M, … • Challenges • Scheduling bandwidth • Scheduling diverse machines and display devices • Standard gateway software • Common software library • Infrastructure with limited life • A view of what a new internet might look like • Driver for the Next Generation Internet Initiative

7. 2.5 Gbps .65 Gbps 40 Gbps 100 Tbps 1.0 Gbps 5.0 Gbps 10 Gbps ESnet Cluster ESnet End-to-end Performance Gap ESnet Host SITE LAN One Host End-to-End Bit Rate Budget for Science Applications Bit-Rate DWDM (One Lambda) Research Opportunities SONET (One Channel) OC-768 OC-192 TCP/IP (One Flow) Apps Site Security Firewall OC-48 GigE ESnet Optical Backbone IP Network Services Host

8. Transmission Control Protocol Internet Protocol (TCP/IP) • Features • Reliable end-to-end transmission control protocol for the Internet • Error control algorithm, CRC-32, optimized for 1.5KB packet size • TCP stack resides in host • “Fair” to all users; Prohibits aggressive use of the network • Constraints on throughput • [1] Congested network: 0-15% of link capacity • [2] Uncongested network: up to 30% of link capacity • Used for FTP • Issues • Maximum throughput ~13 Gbps • Maximum number of packets 2 32 • Designing effective congestion mechanism require router modification • Increasing packet size requires new error control mechanism • Operating system bypass needed to reduce host congestion

9. Firewalls • Features • Monitor, process and filter ports and packets, in real time • Compute intensive • Utilized for cyber protection • Issues • DOE Cyber Security Program Plan Guide, DOE G205.1X (Draft), • 2.3a. Perimeter Protection- “…describe approaches used to secure the perimeter (e.g. firewalls and other intrusion detection systems)”. • Commercially available at OC-12; OC-48 firewalls are rare • OC-192 not in the pipe line

10. Network Environment Trajectory • High Energy Physics • Billion dollar facilities • Multinational research collaborations • Massive data distribution and analysis

11. A Hierarchical Data Grid as Envisioned for the Compact Muon Solenoid Collaboration

12. Climate ModelingProjection for 2008 • Possible features in 2008 • A small number of data repositories; numerous computing sites  • Additional simulation elements/components as understanding increases • 100 TBytes / simulation (100 yr run)1-5 PBytes / yr (NCAR) • Simulation data available to major users for post-simulation analysis • ….and process • Develop fully integrated climate simulation capability, include all high-impct features. • Enable analysis of simulation data by entire collaborating community.

13. Network NeedsVary Widely

14. Major R&D Network Characteristics

15. ESnet Utilization First Cut A Multi-Program Lab October 2002 A Major Computing Center October 2002 Total data =14.7 Tbytes/mo Total data =3.8 Tbytes/mo • Application deduced from “port numbers”. • Any port/application that represents less than 1% of the traffic is labeled “miscellaneous”. • “Encrypted Sessions” are Secure Shell sessions.

16. FTP TRAFFIC PROFILE- SLAC Major activity- ftp file transfers of BaBar data to University of Padua, Italy Ave. Total data =12.6 TB/mo

17. Network Environment Vision

18. Candidate Options Business Model • Growing recognition that TCP/IP-based networks only meets some needs • Advent of wavelength services (DWDM) • Networking slump; so industry may be eager to explore alternatives

19. Observations • Network environment in 2008, dramatically different from the present. • Current description of the network environment fails to capture special nature of Office of Science research endeavors • Data on utilization/performance of ESnet • System: TB/mo., bandwidths, configurations • Local: limited number of sites, data not quantified • Does not meet MICS-program needs (What about users needs?) • Detailed comments (including revised ppt slides) are needed and are welcome. • First opportunity to tell the story- Congressional Hearings March, 2003

20. ESnet Backbone Traffic (1994-2002) An informative “slice” to address MICS program needs

21. Network Transit Times Another “look” to address MICS program needs