Keynote 1: Eliminating Network Delay and Packet Loss Without a Clean Slate
Wednesday, February 22nd, 11:00-11:30am
 Larry Roberts
CEO
Roberts Consulting
Recognized as one of the four founders of the Internet, Larry Roberts received the National Academy of Engineering Draper Award and the Principe de Asturias Award "for the development of the Internet" along with Leonard Kleinrock, Robert Kahn, and Vinton Cerf. Currently an independent consultant, he has founded 4 companies; NetExpress, ATM Systems, Caspian Networks, and Anagran, continuing his goal of improving network technology.
Larry Roberts was the Chief Scientist of the Advanced Research Projects Agency (ARPA) where he designed and built ARPANET, the first packet network and the predecessor of the Internet. While at ARPA, he wrote the first email program and invented the access technique now used for WiFi and wireless. He later founded and ran the first commercial packet network, Telenet (now Sprint), and invented the X.25 protocol which dominated world networking from 1975 to 1990. His other awards include the L.M. Ericsson prize for research in data communications, the ACM SIGCOMM award, the IEEE Internet award, and the NEC Computer and Communication Award. A member of the National Academy of Engineering, he holds BS, MS, and PhD degrees from MIT.
Abstract: Eliminating Network Delay and Packet Loss Without a Clean Slate
Designers and operators often believe that eliminating delay and packet loss in networks would require a complete rebuild or “clean slate”. Since this is virtually impossible to do, the International Telecommunications Union (ITU) has taken a new approach. ITU has recently approved the Q.3313 protocol, which optimizes response time and throughput and eliminates lost packets. Usually, when TCP, UDP, and IPSec traffic traverses Ethernet networks, switches and routers introduce delay and packet loss even under modest loads. TCP, in particular, operates with slow start, causing major delays for interactive traffic while the flows try to increase their speed. The new ITU protocol provides optimal performance for TCP, eliminating slow start and allowing each flow to jump to the maximum rate feasible, considering other traffic, after one round trip. For web access, this would provide sub second page times, a 10:1 improvement. The protocol also eliminates network delay and delay jitter, leaving only speed of light delays. This means the round trip time would be improved typically 3:1. Peak TCP data transfer rates are currently limited by retransmission buffer size to under 100 Mbps, but Q.3313 would immediately permit transfer rates up to 1 Gbps or more, depending on the minimum trunk speed. If deployed in an office complex, Q.3313 could greatly reduce file transfer times and provide fractional second interactive access, thus greatly increasing productivity. In an ISP point of presence, it would improve customer satisfaction dramatically. The presentation will explain the reasons for current network congestion problems and how the new protocol fixes them.
|
Keynote 4: Ethernet-Based Data Centers in the Cloud Computing Era
Wednesday, February 22nd, 2:30-3:00pm
 Greg Sherer
Vice-President - Server and Storage Strategy
Broadcom Corporation
His previous positions include being CTO for Emulex and Neterion. He is a technology industry veteran with almost thirty years of experience in engineering and business development. His diverse career includes contributions as both a technical team member and a proven leader in bringing emerging products and technologies to market. Greg holds several patents and remains active in new product development.
Abstract: Ethernet-Based Data Centers in the Cloud Computing Era
For years, the corporate data center has been growing and evolving by increasing the scale and speed of three separate entities to keep up with IT demand, namely:
- Server farms
- Storage networks
- Data networks
Much like the process by which diamonds are created, data centers are being reshaped dramatically through current economics and the need for business agility. This session will explore the following pressures on data center networks:
- The new edge - server and I/O virtualization
- One wire - network and storage convergence
- Fast, flat, and fat fabrics - network virtualization
These architectural advances, which have been heavily influenced by the advent of cloud computing, will forever change the shape of the data center. The common thread that unites and provisions the new data center is Ethernet. We will discuss how faster speeds and flatter and converged networks that are virtualized at every level (server, I/O, and network) form the foundation of the new data center fabric.
About Broadcom
Broadcom is a major technology innovator and global leader in semiconductors for wired and wireless communications. Broadcom products enable the delivery of voice, video, data, and multimedia to and throughout the home, the office, and the mobile environment. We provide the industry's broadest portfolio of state-of-the-art system-on-a-chip and software solutions to manufacturers of computing and networking equipment, digital entertainment and broadband access products, and mobile devices. These solutions support our core mission: Connecting everything®.
Broadcom, one of the world's largest fabless communications semiconductor companies with 2010 revenue of $6.82 billion, holds more than 15,700 U.S. and foreign patents and applications. It has one of the broadest intellectual property portfolios addressing both wired and wireless transmission of voice, video, data, and multimedia.
A FORTUNE 500® company, Broadcom is headquartered in Irvine, CA, and has offices and research facilities in North America, Asia, and Europe. Broadcom may be contacted at +1.949.926.5000 or at www.broadcom.com . |
Keynote 6: Rethinking the Data Center
Thursday, February 23rd, 11:30am-Noon
Ken Duda
Founder/CTO/VP Software Engineering
Arista Networks
Kenneth Duda is a pioneer in high-performance networking software and lead architect of EOS, a stateful modular operating system for all Arista Networks’ products. He is also the co-author of several network virtualization specifications including VXLAN with VMware and NVGRE with Microsoft. From 2005 to 2008, Ken was the Acting President of Arista.
Before joining Arista, Ken was CTO at There.com, where he played a lead role in designing a real-time 3-D distributed system that scaled to thousands of simultaneous users. Ken was also the first employee of Granite Systems and led the software development effort for the Catalyst 4000 product line after Cisco acquired Granite.
Ken has 3 simultaneous engineering degrees from MIT and holds a Ph.D. in Computer Science from Stanford.
Abstract: Rethinking the Data Center
Network virtualization is a key development in Ethernet network architecture. By decoupling logical topology (as seen by the application or virtual machine) from physical topology (the physical connections between servers and switches), it has a profound impact on network design and usage. Administrators can use it to allow for multipathing, move virtual machines across subnets, stretch virtual LANs between data centers, or provide thousands of secure subnetworks, all without proprietary fabrics or forklift upgrades. This presentation will cover how network virtualization works, what problems it solves, and how equipment vendors and standards bodies are approaching it.
About Arista Networks:
Arista Networks was founded to deliver cloud networking solutions for large data center and computing environments. Arista’s award-winning best-of-breed 10 Gigabit Ethernet switches redefine scalability, robustness, and price–performance. At the core of Arista's platform is EOS, a ground-breaking new software architecture. Arista Networks markets its products worldwide through distribution partners, systems integrators, and resellers. For more information, see: aristanetworks.com |
Saar Gillai
Soni Jiandani
Geng Lin