Editor: Linley
Gwennap In This Issue
If you missed last week’s Linley Tech High-Speed Interconnect seminar, don’t despair! You can download a free copy of the proceedings, featuring presentations from Cisco, AppliedMicro, Freescale, Sarance, Achronix, Luxtera, Xilinx, Finisar, and The Linley Group. Save the date! Mark your calendars for September 16-17 for the inaugural Linley Tech Processor Conference. This two-day event will focus on the newest chips and technologies used in networking and communications applications. Further information will be announced soon. NetLogic Acquiring RMI, Looks to Double in Size Last week, NetLogic announced an agreement to acquire privately held RMI (formerly known as Raza Microelectronics). Under the terms of the deal, NetLogic will pay RMI's shareholders about $175 million in stock plus about $8 million in cash at closing. In parallel, NetLogic is acquiring IDT's network search-engine (TCAM) business. Both deals are expected to close during 3Q09 and would bring NetLogic's headcount to about 500. NetLogic projects the combined businesses will exceed $275 million in 2010 revenue. These figures represent a doubling of NetLogic's team and 2008 revenue, respectively. RMI is shipping several lines of MIPS processors scaling from low-power consumer applications to high-throughput networking designs. The startup generated sales of $79 million in 2008 and $17 million in 1Q09. But during a conference call discussing the NetLogic deal, both companies' CEOs emphasized that RMI's private status had held the company back at Tier One accounts. In particular, RMI has not been successful at Cisco. RMI's largest customer is Dell (EqualLogic), but that design win predates EqualLogic's acquisition. Also during the call, NetLogic CEO Ron Jankov was asked about the possibility of divesting RMI's low-cost Alchemy line. Although the Alchemy products have little in common with other RMI or NetLogic products, Jankov made it clear he intends to retain and invest in the Alchemy line. As a part of NetLogic, RMI should indeed be better positioned against competitors Cavium and Freescale. In addition to sheer size, NetLogic brings strong relationships with Cisco and other Tier One accounts as well as complementary technologies such as its NETL7 regular-expression accelerators. Cavium and Freescale already offer processors with integrated reg-ex acceleration. Although RMI's recently announced XLP832 will rely on a separate accelerator chip, future XLP derivatives could integrate a reg-ex engine. NetLogic can also optimize the interface between XLP processors and future TCAMs using serdes-based interfaces such as Interlaken-LA. Whereas NetLogic's other acquisitions have primarily consolidated the company's position in TCAMs, the RMI deal opens up a new multi-billion dollar market. NetLogic is now poised to expand from leadership in a relative niche to emerging supplier in a far more competitive arena. —Bob Complete coverage of NetLogic and RMI appears in our upcoming report A Guide to Security Processors and Accelerators.
Late last month, Netronome announced the NFP32xx line of network processors. The NFP chips are the first products to come out of Netronome's 2007 licensing agreement with Intel, which enabled the startup to develop chips based on Intel's highly successful IXP network-processor architecture. Due to sample next quarter, the NFP line includes the 20Gbps NFP3240 with 40 microengines and the 10Gbps NFP3216, which has 16 microengines and can serve as a direct replacement for Intel's IXP28xx. The NFP32xx includes an ARM11 control CPU in place of the IXP's ARM-compatible XScale CPU. To support up to 20Gbps of throughput, the NFP32xx includes dual packet interfaces that each support XAUI, 4xSGMII, or Interlaken. The chip uses commodity DDR3 SDRAM for its main memory and optionally supports QDR SRAM and TCAMs with an LA-1 interface. Unlike other 10Gbps+ NPUs, the NFP32xx includes a security engine for bulk crypto and public-key offload. The chip also includes 64-bit hash engines, which can be used to accelerate regular-expression processing. Another unique feature is the NFP's PCI Express interface that supports I/O virtualization, which provides an optimized interface to multicore hosts such as Intel Xeon platforms. While Netronome has been developing the NFP, Intel has been making a renewed thrust into high-speed embedded designs with its Nehalem architecture. Combining the NFP for flow processing with a multicore x86 processor for application processing creates a very powerful and flexible platform for security and L4-7 applications. What Netronome lacks, however, is a clear story on how this approach scales down to higher-volume designs. Still, the NFP32xx is far from just a warmed over Intel design ported to a new process. Netronome has made thoughtful and significant changes to the design to improve performance and integration while reducing system cost and power dissipation. —Bob Complete coverage of Netronome and the NFP323xx appears in our recent report A Guide to Network Processors. Last week, Intel announced that it would acquire Wind River for $884 million. Wind River's VxWorks, a leading embedded operating system, accounts for more than 70% of the company's revenue. In addition to VxWorks, Wind River also offers a version of Linux and various other embedded software. Intel is the leading supplier of general-purpose embedded processors, with a 40% market share. Although rarely seen in deeply embedded designs, Intel's x86 processors are popular for point-of-sale terminals and ATMs, kiosks and signage, and industrial control. Designers select them for their high performance, their associated software, and the availability of ready-made platforms. Intel's stated goal for the acquisition to strengthen its position in the embedded—particularly mobile—market. Wind River already makes its software available on the x86 platform, so it is unclear how the acquisition helps Intel. Wind River offers domain expertise in a few areas, such as wireless basestations and fixed-mobile convergence, where Intel is weak. But overall, the software vendor's major markets, such as industrial and networking, align with the processor maker's strongest areas. Conversely, mobile systems are a small part of Wind River's business. Wind River supports multiple architectures besides x86, including ARM, MIPS, and PowerPC. Intel plans to operate Wind River as an independent subsidiary that will continue to support competing architectures. We expect this support to wane over time, however, and other processor vendors will probably seek alternate software partners that are not owned by their leading competitor. Thus, the popularity of VxWorks, already under attack from the less-expensive Linux, is likely to decline. The Wind River deal is the first large transaction for Intel under the leadership of Paul Otellini. Otellini took the helm of Intel in 2005, following Craig Barrett. Barrett had made several major acquisitions, the signature attribute of which was their dismal outcome. It is, therefore, unsurprising that Otellini has not been as acquisitive as Barrett, but it is hard to see how the Wind River deal will have a positive outcome for Intel. —Joe Additional coverage of Intel's embedded processors appears in our report A Guide to High-Speed Embedded Processors.
At Interop, Marvell announced its Prestera 98CX8234 packet processor for data center applications. The device is the first in the industry to offer 48 x 10GbE ports and the first to integrate 40GbE MACs. The company offers different variants of its 98CX devices, which use the 48 ports to realize different combinations of 10GbE and 40GbE ports. In addition to the largest number of 10GbE ports, Marvell has added the data center bridging features that are under definition in IEEE 802.1 task groups. To support Fibre Channel over Ethernet (FCoE), the 98CX8234 supports FCoE initialization protocol packets, which are used to discover FC entities, initialize ports, and forward FC traffic. Marvell was late to enter the market for 10GbE switches, and in the past, its switches lagged on the number of 10GbE ports. With the 98CX8234, the company has leap-frogged competitors to lead on port density, 40GbE, and converged data-center features. Because of the number of new features, the biggest concern is the production level functionality of the device. After announcing the Prestera CX in November, Marvell is now sampling that device. If the company can quickly qualify the 98CX family for production, we expect several OEMs to select the 98CX8234 for data-center switches. —Jag Additional coverage of Marvell's Ethernet products appears in our report A Guide to Ethernet Switch and PHY Chips.
The network-security market remains as dynamic as the threats that are driving its growth. Security-equipment and software vendors face increasing requirements from their customers who need intrusion detection and prevention from ever-increasing threats. In most cases, standard processors alone can no longer deliver the required performance in a cost-effective and power-efficient manner. To meet these challenges, a new breed of processors is taking integration to the next level by integrating one or more CPUs, memory and I/O controllers, and special-purpose engines for security functions. A Guide to Security Processors and Accelerators covers processors that integrate high-throughput encryption, such as Cavium's Octeon families, RMI's XLR/XLS/XLP, and Freescale’s MPC8572 and P4080. We also cover unique processors from Netronome and Tilera, which integrate 40 or more cores and target network-security applications. In addition, the report examines vendors developing content-inspection accelerators: LSI, NetLogic, Lionic, and cPacket. Finally, we cover IPSec and SSL accelerators from Cavium, Exar (Hifn), SafeNet, and Broadcom. With one report, you can quickly compare the key vendors and their products and accelerate your selection process. The report analyzes each vendor and each product, probing their strengths and weaknesses and presenting key details in a consistent, easy to compare fashion. For example, we have sorted through various performance claims to put all the security processors on a level playing field, showing who can deliver real system throughput. We also tell you who is really delivering on packet processing claims and who is just delivering marketing hype. Order by July 17 to take advantage of the prepublication price. For further details, visit our web site.
To receive Linley Wire via e-mail, click here
|
