• Technical Conference:  30 March – 03 April 2025
  • Exhibition: 01 – 03 April 2025
  • Moscone Center, San Francisco, California, USA

SC359 - Networking for Datacenters and Machine Learning

Sunday, 24 March
09:00 - 12:00 (Pacific Time (US & Canada), UTC - 08:00)

Short Course Level: Beginner


Hong Liu, Google, USA
Ryohei Urata, Google, USA

Short Course Description:

Over the past decade, datacenters have rapidly grown as an enabler of new applications and capabilities (cloud computing, machine learning, internet content streaming, internet search, social networking, etc.). As such, networks inside the datacenter have become the singular force driving the business and technology development of corresponding fiber optic communications.

This course describes various network architectures and technology considerations in constructing datacenter networks for traditional as well as burgeoning machine learning applications. Particular emphasis is placed on the role of optical interconnects, for which we discuss technology trends, key metrics, the progression to higher interconnect speeds, and roadmap for the next 3 to 4 years. The role of optical switching is also discussed.

Short Course Benefits:

This course should enable you to:

  • Define warehouse-scale computers (WSCs) and describe their structure
  • Describe the engineering principles and philosophies in building scalable mega-datacenter infrastructures
  • Compare different datacenter network architectures/topologies and understand their tradeoffs
  • Understand key electrical technology building blocks (switch silicon, backplanes, copper interconnect) for datacenter networks
  • ​Identify key trends and metrics for intra-datacenter optical interconnects
  • Select suitable optical/optoelectronic interconnect technologies for building datacenter networks and machine learning systems
  • Understand relevant optical interconnect standards/multi-source agreements (MSAs) across multiple generations ( 100GbE, 200GbE, 400GbE)
  • Explain the emerging roles of optical switching
Short Course Audience:

This course is beneficial to optoelectronic engineers, fiber optic transceiver designers, and optical transmission engineers who would like to understand the requirements of datacenter and machine learning networking. It would also benefit network engineers looking for deeper knowledge of high-speed optical communication technologies used to realize various datacenter network applications. For network planners and architects, this course provides outlooks in optical network technology development for the next 3 to 4 years.


Instructor Biography:

Ryohei Urata is currently a Principal Engineer/Director in the Platforms Infrastructure Engineering (PIE) Optics Group, where he has defined/developed Google's datacenter optical technologies and corresponding roadmap for the past decade. Prior to joining Google, he was a research specialist at NTT Photonics Laboratories, Japan. He has over 150 patents, publications, and presentations in the areas of optical interconnect, switching, and networking. He received a Ph.D. degree in electrical engineering from Stanford University (Stanford Graduate Fellow). He was elected an Optica/OSA Fellow in 2022.

Hong Liu is a Google Fellow at Machine Learning, Systems and Cloud AI (MSCA), where she is involved in the system architecture and interconnect for a large-scale computing platform. Her research interests include high speed signaling, network architecture and optical interconnection for cloud and machine learning. Prior to joining Google, Hong was a Member of Technical Staff at Juniper Networks, where she worked on the architecture and design of network core routers and multi-chassis switches. She holds 55 U.S. patents in the areas of photonic devices, datacenter networks, and optical communications. Hong received her Ph.D in electrical engineering from Stanford University, and is an Optica Fellow.