next up previous
Next: About this document Up: SuperComputing-94 --- Networking Technologies Previous: The Future of

Conclusions

Supercomputer networking has a broader ranger of requirements than general corporate enterprise networking. While the requirements for user access to workstations via (X-windows-based) client/server applications exist in both networking communities, the supercomputing community has requirements for significantly greater network bandwidths to support clustered computing applications and massive input/output requirements. To support supercomputing data requirements in the bus environment, SCI networking can provide latency at gigabit-per-second network speeds to support shared memory applications on a workstation cluster. Serial HiPPI and Fibre Channel are competing to provide switched gigabit-per-second networking capabilities for the I/O and channel environments. ATM offers the potential for seamless networking within the LAN, MAN, and WAN environments. It is important to remember that all of these technologies are still immature. They each appear technically promising, but will all of these technologies have the financial backing to mature? Will there be a demand for SCI technology as more capable workstation clusters are developed? Will both HiPPI and Fibre Channel survive? Will standards-based ATM user-network interface (UNI) and network-network interface (NNI) protocols be implemented for true multi-vendor interoperability? Will the cost for ATM interfaces and ATM network services drop to a point where ATM will be the user network of choice?

My opinion is that all of these technologies will survive, but various technologies will be effected by different market forces. SCI, HiPPI, and Fibre Channel will most likely survive as vertical market items --- high cost specialty components used in a limited market. While these technologies survive and prosper, their cost/bandwidth ratio will remain high, and only limited numbers of companies will be able to generate revenue with these technologies. Meanwhile, ATM will expand its market share until this technology commands a horizontal market status. ATM will be massed produced to support the telecommunications and video-entertainment industry, and the benefits of inexpensive custom circuitry will be available to the computer networking community. A significant problem that ATM must face is the inability to develop the technology rapidly enough to support the massive thirst for bandwidth that is developing within world-wide corporate enterprise networks, the National Information Infrastructure (NII), and the Worldwide Information Infrastructure (WII).

When comparing the opinions of the highly technical members of the Gigabit Networking Round-table to the more secular members of the Future of the Internet Panel, opinions often were nearly contradictory. There was little disagreement between the technical members of the roundtable, or between the members of the panel; however, there was significant differences between groups. The opinions of technical people were that networking technology still had far to advance in order to be able to proliferate throughout our day-to-day lives. Meanwhile, those people involved in government policy areas declared that the technology issues were of less importance than the social issues of using the technology as the means to empower all Americans and provide equal opportunity for advancement. Technically oriented people expressed the view that there must be continued revolutionary improvements before there will be the scalable network technology available to provide the bandwidth to support true multimedia applications into all homes. The non-techies expressed the concern that the applications do not exist today that will permit the majority of Americans to have a positive experience from Internet connectivity. I believe that the hardware issues can be solved for cost effective large-scale proliferation of bandwidth in the 2000-2020 time frame, but I am less confident that the applications issue can be resolved because it will require overcoming many problems that exist at the most difficult of all levels, the user-machine interface.



next up previous
Next: About this document Up: SuperComputing-94 --- Networking Technologies Previous: The Future of



David P. Koester
Sun Oct 22 13:05:27 EDT 1995