Banderacom, a 10Gb InfiniBand chip startup, has been restructured as NetEffect “to develop silicon to enable the next generation of high-performance, multi-gigabit Ethernet.” NetEffect has secured $22 million in Series A financing led by TL Ventures and joined by new investors Granite Ventures, TI Ventures and Duchossois Technology Partners, along with existing investors Austin Ventures, JatoTech Ventures and Infinity Capital. NetEffect has 25 employees, including eighteen former Banderacom employees.

Senior members of the technical staff at Jato Technologies, a networking chip startup that was acquired by Intel, founded Banderacom was in November 1999 to develop InfiniBand chips. Bandercom had raised $9M in Series A funding from Austin Ventures, Jatotech and Intel, and Series B funding of $35M in June 2001 from Infinity, Trinity, and Qlogic. In late 2002, the decision was made to suspend InfiniBand chip development efforts to focus on the emerging TOE/RDMA Ethernet NIC market. NetEffect retained Banderacom’s VP of Engineering, as well as the architects and designers who brought Banderacom’s 10G chips to production. NetEffect has a proven silicon team and will re-use Bandercom’s IP and architectures.

According to NetEffect, data center servers will require 10Gb connectivity by 2006. However, realizing 10Gb and even multiple Gb performance requires the elimination of overhead caused by TCP/IP processing, intermediate buffer copying and application context switches. The migration from 100Mb to 1Gb increased server overhead 10x, resulting in unacceptable levels of server CPU processing and memory bandwidth consumption. And the migrating to 10Gb will increase server overhead 10x yet again.

NetEffect lab test data indicates that 40% of CPU overhead results from transport processing, 20% from intermediate buffer copying, and 40% from application context switches. Transport Offload Engines (TOE) move transport processor cycles to the NIC, and move TCP/IP protocol stack buffer copies from system memory to the NIC memory. RDMA/DDP eliminates intermediate and application buffer copies, reducing memory bandwidth consumption. And Kernal bypass (direct user-level access to hardware) reduces application context switches. A TOE-enabled, iWARP-enabled networking stack would eliminate transport overhead, buffer copies, and context switches.

NetEffect ICs “allow Ethernet to easily scale to meet data center demands, using next-generation iWARP RDMA Ethernet that is fully compatible with the Ethernet infrastructure found in today’s data centers.” Today’s servers require multiple adapters to connect to Storage, Network, and Cluster Fabrics. iWARP, which is compatible with today’s Ethernet infrastructure, virtually eliminates networking overhead, enabling the delivery of multi-gigabit Ethernet and convergence of data center traffic for networking, storage, and clustering over a single network technology.

Servers with the NetEffect iNIC deliver optimized file and block storage, networking, and clustering from a single adapter. IT managers can support all data center fabrics with just Ethernet. For blade servers, the iNIC reduces I/O real estate, power, cost, and SKU options while delivering increased flexibility.

Competitors include Alacritech, Chelsio, Ammasso, Intel andS2io. Neteffect believes that its patented virtual pipeline architecture, which reduces latency, provides it with a distinct competitive advantage. The company plans to have first silicon later this year and is currently choosing a foundry partner.

Rick Maule, CEO, President & Chairman (previously VP and GM of the Mobile Communications Division of 3Com)

Terry Hulett, VP of Engineering (previously VP of engineering at Banderacom)

Rob Senders, CFO (previously CFO for Banderacom and VP of finance for Compaq’s Industry Standard Servers Group)

www.rdmaconsortium.org