30-month Program Review & Network Demonstration,

September 16/17, 1997, LLNL

This was meeting to review the status of the NTON in particular since its funding runs out in the near future. They are looking to get funding extended for another 2 years. They are particularly interested in real driving applications now that they have a network infrastructure in place. My purpose was to expose attendees to BaBar's WAN needs. There were about 50 attendees, with representatives from DoD, DoE, LLNL, carriers such as PacBell, component manufacturers such as Nortel, Rockwell, current and potential application interests (e.g. Kaiser, SLAC, Sandia).

I talked to Bob Aiken of DOE, who was at the meeting, about DOE’s interest and SLAC’s needs. He asked for a copy of our transparencies. He also mentioned the money DOE/MICS is putting up for network research. It is unclear how much will be available, to some extent it depends on how much DOE eventually gets from NGI. Currently there is $10M of redirected money. Bob expects to make 2 or 3 grants from a few hundred thousand dollars to a couple million. He is not looming for research needs requiring simple raw bandwidth since that is available (at a cost) commercially. He is looking for applications that require network research. He mentioned two examples. The first would be to use NTON and make dynamic automated bandwidth reservations or quality of service for different applications, such as for a video conference for low packet loss for an hour or so. The second would be to provide security over a WAN for say an object oriented data base, maybe at the object level. The latter would not just simply be using a CA/KDF (since that is commercially available). The work could be done in conjunction with a vendor (e.g. Objectivity) but should not simply be testing beta code from the vendor.

Goals:

• Advanced networking technologies, QoS & Security
• High Connectivity, ultra hig speeds,
• Advanced applications

SuperNet (Ultra high speed goal)

• Broadband trunking & local exchange technology

• WDM MAN rings
• Want to link WDM MAN to form WAN
• Broadband ( >= DS3) access for high end users, sharing the same fiber facilities with common users, could be using statistical multiplexing, delivering analog, digital, and video applications, with IP over WDM, understand the architecture, hardware and software.

Ultra fast packet switching & networking technology (Tbts/sec)

• Distribute the switching in order to get arund requiring single big switch which will not scale, with some kind of WDS/ATM
• Looking at Soliton technology

Addressing promise of building scalable, modular, survivable networks using optical switching. Now have a fully functional 4 wavelength (each with 2.5Gbps), all-optical, 500km bi-directional (daul counter-rotating) rings using ATM-SONET integrated optical layer integrated, with wavelength switching, routing and re-use. Includes nodes at LBL, LLNL, PacBell San Ramon, Sprint Burligame. It is being operated as an open network to allow its use as a testbed for WDM technologies and applications. Ports available for user connections are 28*OC3, 6*OC12 ports

The team includes: Nortel, LLNL, UCSD, Hughes, Sprint and Pac Bell with associate members Uniphase, Rockwell, Case Western Reserve U, Columbia U.

Had first on the floor demonstration at San Jose show Feb-96.

Issues:

• Dynamic power balancing critical to optical switching
• WDM nets v. sensitive to gain at EDFAs
• NTON deployment mirrors real worls evolution
• Only equipment that can be manged will be introduced into the netowk
• Only networks that are survivable will be deployed

Network management is a challenge, standards available for existing technologies (SONET/ATM). Optical network management standards are just emerging, management is required within and across layers. Have a prototype management system that covers network elements, network controllers and the network layer. The layer interactions are not well understood, and continued research & development is crucial to commercial network development.

Bandwidth costs to fall by factor 10 over next 6-10 years (e.g. $1M/yr for 155 Mbps today, ~$1M/yr for 1.5Gbps by about 2005. Depends on large growth in traffic volume & rapid evolution of technology for coarse bandwidth management, > 100Gbps per fiber, increased optical reach (to 1000km).

Bay Area is "target rich" for research etc and people to understand how to use WDM, companies like Kaiser, Lockheed, 3Com, Xerox Su, Cisco, establishments such as SRI, Stanford, LLNL, LBL, UCB, SNL & NASA have applications and needs that can drive

the deployment.

Short term focus areas: activate an OC-48 link yo NASA Ames, expand network access links to reach additional users (particularly the south Bay), conduct wavelength switching & power balancing experiments, integrated other DARPA (&non DARPA) sponsored technology into network. NTON will continue to operate the network & support user applications, will provide modest levels of equipment necessary to connect user applications, users will need to provide their own resources to distribute the wavelength at their site. Want to position network as transport platform for NGI research.

Users connect via direct connect thru:

• local fiber plant at LLNL, Sandia, UCB, LBL, Sprint, ATL, and NERSC;
• direct connect thru PacBell fiber plant (ICSI, Chevron);
• tie to other nets (SRI (OC3) via Sprint’s Silicon Valley Test Track);
• and SONET tributary multiplexing 1310nm signal over backbone fiber with OC3 tributaries using Pac Bell fiber plant UCSF, Kaiser SF, Kaiser DOR, Kaiser Macarthur.

Most (12) sites are OC3, LBNL, LLNL & SNL have OC12.

Main issues have been controlling the capability, diagnostics are needed, also need ways to manage the control and diagnostics (maybe using AI), plus modeling tools.

A major gating factor for introducing new technology is managing it, another factor is how to roll it in to an existing operational networking. With a large (Pac Bell) network they would have roughly 20% error rate in a static configuration database, so need to get the information dynamically by self discovery.

SLAC identified as a new Research Partner Application site. Will start working next week to bring in fiber from Pac Bell (assuming NTON goes ahead) into new sites.

IP is carried over ATM & SONET. Can also transport analog & digital TV. Tow of 4 wavlengths are dedicated to SONET traffic, this is what will be used by most of the applications. Not a 24x7 network (next business day), but it stays pretty stable except when changes are made. There is some redundancy in particular at the ATM layer, and of course IP recovers from problems.

Proposal to DARPA that has been accepted. Need platform to do the research, have selected CAIRN as the testbed. Collaborative Advanced Interagency Research Network = CAIRN. Propose connection (hopefully soon) possibly via LBL. Areas to be explored include: multicast, RSVP, Ipv6, management. Collaboration is mainly with agencies and universities.

Real-time applications have critical network requirements. Includes QoS. The core infrastructure technologies are connection-less IP with connection oriented ATM up to 622 Mbps. There are challenges to how to integrate the technologies. Overlay of IP over connection ATM with transition between fixed size ATM cell and larger IP packets. Adapting solution from Ipsilon - high speed router with ATM as switching fabric. Default hop-by-hop for short-lived (IP) flows, set up ATM SVCs for longer term flows. Traffic characterization allows the generation of realistic traffic models. Models based on existing empirical data. They will use NTON as the test-bed working with NASA Ames, SNL, LLNL, UCB and the ACTS satellite. They have an OC12c long link emulator that can program bit-error rates, burst size, intrarrival times, selectable delays (up to 1.4s).

Tektronix has a new video network interface division. Looking at compression, Internet pull.

Want to service video content production services (i.e. moving it around and making it useful (e.g. editing in real time) to the customer). Looking at MPEG2 & MPEG4, looking for people with requirements. Video they are talking about 270Mbps digital uncompressed video streams. This is put into AL0 or AAL5 ATM interfaces with one or two video signals to independent VPI/VCI on OC-12 interface. Has clock recovery. They also have an internally developed an OC-48 network interface.

I gave a presentation on BaBar’s WAN requirements. The requirements include:

• 20Mbytes/sec to transfer reconstructed data to other sites such as LBNL;
• interactive event analysis by people at remote sites requiring several Mbytes/sec per user and expect several simultaneous users;
• program builds between sites, requires copying of 200 Mbytes/user in 20-60 seconds;
• and high quality collaboratory work with video conferencing (say 5-20 simultaneous conferences or 4Mbytes/sec for long (>~ hour) periods, remote real-time data monitoring and experiment feedback, remote presence at the experiment (full motion video with several Mbits/sec), evaluating the effectiveness of virtual experiment / reality etc. for HEP (build on LBNL setup).

Future research for consortium, the technology they have in place, things which look interesting.

Extend the network reach. A Teraflop testbed linking all California supercomputing centers, mass memory projects and major universities, plus connections up to Seattle.

Enhance the infrastructure. Capable of 32 wavelengths and OC-192 per wavelength for increased capacity. 8x8 è 16x16 cross-connects, more nodes in Bay Area for increased network flexibility. How are new users connected up to backboe, linear, ring, what kind of redundancy.

Look at architecture issues. How to bring Terabit to reality for backbone, what does it mean to switching and to protocols. What are effective access multiwave protocols and how do they help resolve problems not addressed by current optical networks.

Virtual networks using "dark" wavelengths, high performance computer interconnects, "Grand Challenge" computing support.