PROCESSORS

Siliquent Technologies emerged from stealth mode, outlining its plans to bring a new product category to market called an "Ethernet Processing Unit" (EPU) -- a new kind of Ethernet coprocessor that enables Ethernet to extend into high-performance applications for the first time. The company's new coprocessors and related software will solve the application-processing bottlenecks that have prevented Ethernet from moving into applications such as storage, distributed databases, and cluster computing. "The industry has a long-held vision of a single ubiquitous networking interface for networking, storage, and computing," said Charles Chi, president and CEO of Siliquent. "The ecosystem of applications, protocols, and programming interfaces is in place for Ethernet to become that 'unified wire.' However, application-processing bottlenecks have limited Ethernet's pervasiveness. Siliquent's innovation will help overcome that bottleneck and move the industry closer to fulfilling the vision for one protocol in the data center." As a fabless semiconductor company, Siliquent will sell its EPUs and software to a variety of OEMs which satisfy a breadth of networking applications. Siliquent's products will be incorporated onto network interface cards (NICs). As a result, a new category of "intelligent" NICs (iNICs) will emerge that provide a "unified wire" -- standard Ethernet operation for general-purpose networking, and also for high-performance I/O applications. Moreover, because the coprocessors will be simultaneously able to support multiple upper layer protocols, they can be used in a variety of applications, offering greater deployment flexibility for OEMs. Ethernet's unrivaled ubiquity and economic advantage, combined with its 1 Gbps and 10 Gbps data rates, make it a compelling choice for high-performance networking applications. With that goal in mind, the industry has developed a number of protocols for supporting high-performance applications over Ethernet. These include iWARP (RDMA over TCP/IP) and iSCSI (block storage over TCP/IP). These protocols support the large installed base of software in high-performance applications through support for the most pervasive APIs, including kDAPL for NFS, uDAPL, Message Passing Interface (MPI), and Sockets Direct Protocol (SDP). Despite having the protocols and APIs in place, applications beyond general networking have been out of Ethernet's reach due to limited application throughput. Until now, application throughput has remained an issue because traditional architectures require the host CPU to perform all tasks related to processing Ethernet communications. Host CPU cycles are consumed for processing the transport protocol, the upper layer protocols, and moving data within memory space from the kernel to the applications. With modern communications-intensive applications and multigigabit data rates, few CPU cycles remain for processing the actual applications. "There is great demand in the high-performance computing arena to significantly improve applications-processing throughput over Ethernet," said Arun Taneja, founder of the Taneja Group. "Siliquent's technology is a well thought-out and innovative approach that has the potential to drive the proliferation of Ethernet in high-performance applications." Siliquent's EPUs and software are currently in evaluation at high-profile server, storage, and computing system and subsystem OEMs. The company plans to announce specific details of its multigigabit, multiprotocol EPUs later this year. "With chips already in the hands of our customers, we are now the first company to deliver iWARP-compliant chips which perform at 10G wire speed," said Amit Oren, chief technology officer and a founder of Siliquent. "This performance, combined with extensive support for existing standards and software applications, should drive the broad deployment of high-performance networks based on Ethernet, and move the industry closer to a single unified-wire technology for the data center."