Draft:Tiny Tera

  • Comment: is widely recognized as one of the first terabit-class switch prototypes and a landmark in the evolution of high-speed Internet routing. If true, this would require more substantial justification than the references given. I prefered the previous lead of the article more; the current lead is very promotional, not encyclopedic. Caleb Stanford (talk) 20:46, 31 August 2025 (UTC)

Tiny Tera is widely recognized as one of the first terabit-class switch prototypes and a landmark in the evolution of high-speed Internet routing. Developed in the late 1990s at Stanford University in collaboration with Texas Instruments, it showed that high-performance router cores could be built using commodity CMOS technology rather than specialized optical hardware. The prototype used 32 ports at 10 Gbit/s each for an aggregate bandwidth of 320 Gbit/s, and became a widely cited case study in the design of scalable high-speed switches.[1]

Overview

[edit]

Tiny Tera was designed to overcome limitations in router backplanes and memory bandwidth at a time when Internet traffic was growing rapidly. The system used an input-queued switch architecture and a method known as virtual output queuing to mitigate head-of-line blocking, a long-standing problem that reduces throughput in queued switches.[1] A modular “sliced crossbar” design allowed the switch to operate in parallel across many one-bit-wide slices, while a centralized scheduler configured the crossbar every packet time.

A key innovation was the use of the iSLIP scheduling algorithm, a round-robin arbitration method that achieved high throughput and fairness while being simple enough for hardware implementation.[2]

Architecture

[edit]

Tiny Tera’s main architectural features included:[1]

  • A 32×32 crossbar core implemented as multiple one-bit slices operating in parallel
  • Input buffers organized into per-output queues (virtual output queues)
  • A centralized scheduler running the iSLIP arbitration algorithm to efficiently match inputs to outputs each cycle[2]
  • Support for both unicast and multicast traffic, including fanout-splitting to reduce blocking
  • High-speed serial chip-to-chip links to interconnect port cards and the switch core

These design choices allowed the prototype to scale bandwidth while remaining manufacturable with commodity CMOS parts.

Technical Significance

[edit]

Tiny Tera demonstrated that a switch core capable of hundreds of gigabits per second could be built with commercially available technology. At the time, most high-end routers were limited by memory or bus bandwidth; Tiny Tera showed that scalable fabrics with efficient scheduling could overcome these limits.[1]

Networking textbooks and reference works have cited Tiny Tera as a pioneering example of packet switch architecture:

  • Principles and Practices of Interconnection Networks by William Dally and Brian Towles includes detailed case studies on Tiny Tera’s bit-sliced design and its allocator, situating it alongside industrial designs such as the Cray T3D and IBM Colony router.[3]
  • Advanced Router Architectures by Axel Kloth discusses Tiny Tera in the context of scalable router fabrics and high-availability Internet backbones.[4]

These independent sources identify Tiny Tera as a reference design that influenced both academic research and later commercial systems, including large-scale Internet routers from companies such as Cisco and Avici.

Contributors

[edit]

The project was led by Stanford professor Nick McKeown, with graduate students Martin Izzard, Adisak Mekkittikul, William Ellersick, and Mark Horowitz contributing to architecture and implementation.[1] Researchers from Texas Instruments’ Communications Lab, including Helen Chang and Ani Anirudhan, collaborated on hardware components and high-speed serial link technology.[1]

Legacy

[edit]

Tiny Tera directly influenced subsequent projects at Stanford such as the Stanford Packet Switch and the NetFPGA platform for networking research. Its scheduler design remains a standard reference in computer networking curricula, and the system is frequently cited in academic and industrial literature on crossbar fabrics, switch arbitration, and Internet router scalability.[3]

See also

[edit]

References

[edit]
  1. ^ a b c d e f McKeown, Nick; Izzard, Martin; Mekkittikul, Adisak; Ellersick, William; Horowitz, Mark (January–February 1997). "The Tiny Tera: A Packet Switch Core". IEEE Micro. 17 (1): 26–33. doi:10.1109/40.566194. ISSN 0272-1732.
  2. ^ a b McKeown, Nick; Mekkittikul, Adisak (January–February 1999). "Designing and Implementing a Fast Crossbar Scheduler". IEEE Micro. 19 (1): 58–67. doi:10.1109/40.748793. ISSN 0272-1732.
  3. ^ a b Dally, William J.; Towles, Brian (2004). Principles and Practices of Interconnection Networks. San Francisco: Morgan Kaufmann. pp. 155–157, 383–385. ISBN 978-0-12-200751-4.
  4. ^ Kloth, Axel (2006). Advanced Router Architectures. Boca Raton: CRC Press. pp. 215–220. ISBN 978-0-8493-3550-1.