The
Linley Wire
Independent
Analysis of the Networking-Silicon Industry
Volume 10, Issue 4
March 9,
2010 |
 |
Editor: Linley Gwennap
Contributors: Bob Wheeler, Jag
Bolaria, Joseph Byrne
In This Issue
A Guide to CPU Cores and Processor IP will be available soon with our latest analysis on CPU, DSP, GPU and video-engine IP from ARM, Ceva, Imagination Technologies, MIPS, NXP, On2 (Hantro), Tensilica, VeriSilicon (ZSP), Virage Logic (ARC), Vivante, and more.
Broadcom First with 40nm 10G PHY
In February, Broadcom became the first company to sample a 10GBase-T PHY built in 40nm technology: the BCM84833/4. Although Aquantia was the first company to announce a 40nm 10GBase-T PHY and the startup has developed 40nm test chips, Broadcom appears to have upped all others on availability of first samples. The 40nm node is an important milestone, because this technology reduces 10GBase-T power to tolerable levels.
Broadcom's new PHYs are offered in dual-port configurations for NICs and in quad-port configurations for switches. At 4W per port, the power dissipation of these devices is a bit more than Aquantia estimates for its devices, but the startup has yet to measure actual numbers. Like other 10GBase-T PHYs, Broadcom's parts include XAUI and XFI ports for direct connection to an optical module as well copper cabling. The devices also support auto-negotiation to lower Ethernet rates and thus can be rolled into an installed base of GbE and FE switches.
By sampling 40nm 10GBase-T PHYs, Broadcom has made it very difficult for startups such as Aquantia and Teranetics to win designs at large OEMs. If Broadcom can demonstrate that its new parts meet the 10GBase-T specifications and smoothly move them into production, it could relegate the startups to competing only at Tier-2 accounts, which are not large enough to sustain them. Thus, 2010 promises to be a telling year for determining the 10GBase-T winners among the leading six (Aquantia, Broadcom, Marvell, Plato, Solarflare, Teranetics) contenders. —Jag
Additional coverage of 10GBase-T products appears in our report A Guide to Ethernet Switch and PHY Chips.
Vitesse Updates Ethernet Line
Vitesse has announced a new family of Carrier Ethernet products that the company plans to sample in 2Q10. The product line includes two switches and two MACs. The lead device in this product line is the Jaguar switch, which has 24 GbE ports and four 10GbE ports. Lynx is a derivative version that has 12 GbE ports and two 10GbE ports. The MAC products match the port density of the switches and are pin compatible with the respective switch. These product target Carrier Ethernet applications such as PON MDUs, mobile backhaul, and Ethernet transport and services over the telecom network.
With its detailed traffic-management capabilities, Jaguar enables providers to deploy services such as E-Line and E-LAN. It integrates hardware to support transport protocol options such as PB, PBB-TE, MPLS-TP, and PWE. Jaguar integrates a 32Mb shared buffer for 8 per-port queues with 8 priorities each. The queues can be scheduled using a combination of strict priority and DWRR (deficit weighted round robin) algorithms. Additionally, Jaguar can shape traffic for up to 512 tunnels.
These devices support the timing and OAM (operation and maintenance) features necessary for mobile backhaul and managing connections over third-party networks. For OAM and managing the connection, these devices support IEEE 802.1ag and ITU's Y.1731 recommendation. To meet the telecom's network's protection requirements, they support redundancy options such as 1:1 and 1:n and can perform protection switching in less than 50ms. The devices are rated for commercial and industrial temperatures, allowing them to be used in harsh outdoor environments. Jaguar integrates a 400MHz MIPS24KEc CPU. This CPU can handle control-plane messages, avoiding the need for an external processor.
Vitesse has shifted the focus of its Ethernet products to address Carrier Ethernet requirements. The company claims to be compliant with the MEF's traffic management and Ethernet service specifications. For access switches, Vitesse's CE product line offers a large number of queues and the requisite traffic-management capabilities in a single chip, which should save development time and system cost compared with standard Ethernet switches and FPGAs. —Jag
Additional coverage of Vitesse's Ethernet products appears in our report A Guide to Ethernet Switch and PHY Chips.
EEMBC Releases TCP/IP Benchmark
The industry consortium EEMBC is known for its synthetic benchmarks that measure CPU performance. The group's newest benchmark operates at the system level to measure TCP/IP throughput. This approach goes beyond the CPU, taking into account the memory and I/O performance needed to achieve peak efficiency. Like all EEMBC benchmarks, this one was developed by a multivendor working group.
The new benchmark suite, known as ETCP, uses an Ixia test platform to generate simulated network traffic in real time. This traffic is processed by the device under test, which can be a CPU or NPU on a development board or in a complete system. The test is scalable from 10/100 Ethernet to 10G Ethernet and thus can be run on a variety of devices ranging from microcontrollers to high-end processors. Although the test is referred to as a TCP test, various components of the suite validate performance at Layers 2 through 7.
Comparing the performance of complex multicore processors is quite challenging, because the various architectures used can have a wide range of effects on software performance. Synthetic benchmarks can be helpful, but a system-level benchmark measures actual throughput and can thus point out cases where an implementation has a particular bottleneck. We encourage vendors to use these system-level benchmarks to better prove the deliverable performance of their processors. —Linley
Additional coverage of EEBMC benchmarks appears in our report A Guide to High-Speed Embedded Processors.
News In Brief
Phyworks has announced the availability of its 10Gbps equalizer chip, the PHY2210. The PHY2210 can improve performance in active cable and backplane applications. In data centers, the device is useful for intra-rack, rack-to-rack, and intra-switch applications. With an operational range of 1Gbps to 10.5Gbps, the PHY2210 can be used with SFP+ cables in Ethernet, Fibre Channel, and InfiniBand designs. In addition to the receive equalizer, the chip also integrates transmit pre-emphasis. At 10Gbps, the PHY2210 enables lengths of up to 15 meters using 24AWG twin ax cables. With a small size of 5mmx5mm and a power dissipation of 500mW, the PHY2210 can easily be integrated into the cable connectors. —Jag
Additional coverage of Phyworks appears in our report A Guide to Ethernet Switch and PHY Chips.
Linley Tech Spring Conference Coming May 18-19
What do trends like cloud computing, virtualization, and network convergence mean to your next design? Beyond buzzwords, these trends have very real implications for all types of data-center networking systems and servers. For help sorting all this out, mark your calendar for the inaugural Linley Tech Spring Conference coming to San Jose on May 18-19.
This two-day single-track event is the only one of its kind focused on the processors, components, and interconnects for data-center networking and security designs. The conference will feature in-depth presentations from technology leaders on the newest chips and technologies for switching, server connectivity, and security processing.
The Linley Group will open the conference with an overview of the market, technologies, equipment-design, and silicon trends for designers of data-center networking and security equipment. The remainder of the program will include talks and panel discussions covering a broad range of topics including the following:
- 10GbE switching and backplanes
- Data-Center Bridging (DCB) and 40G Ethernet
- LAN/SAN convergence and FCoE
- Low-latency system interconnects
- Network security
- Application switching
The conference is intended for network-equipment vendors, server OEMs, system designers, network service providers, enterprise-network managers, software developers, press, and the financial community. Get the information you need to jumpstart your design! Attendance is free to qualified attendees. Program details will be announced soon, visit our web site details.
Conference sponsors include AppliedMicro, LSI, Netronome, NetLogic Microsystems, Freescale Semiconductor, Xelerated, Cavium Networks, Intel, Chelsio Communications, the Ethernet Alliance, and the HyperTransport Consortium.
To
receive Linley Wire via e-mail, click
here
About
Linley Wire
© 2002-2010 The Linley Group |