RSS .92| RSS 2.0| ATOM 0.3
  • Home
  • Community
  •  

    HP paper: Killer Fabrics for Scalable Datacenters

    April 20th, 2009

    One of our Wall Street friends pointed us to a paper recently published by HP Labs, which describes why the enhancements we’ve made to Ethernet for scaling are so important in the datacenter.  The team at HP Labs conducted tests on our first-generation FocalPoint 10GbE switch chips; some of the results are summarized in the paper.  Their testing was conducted on our original FM2000-series silicon, and our current-generation FM4000-series silicon offers significant enhancements to hash efficiency, multipath routing, and fat-tree scaling.  Nonetheless, they found that our hash-based symmetric routing is a practical approach for managing large-scale datacenter networks.  The following is the abstract for the paper:

    Large-scale datacenters are rapidly increasing in number and size to satisfy computing and storage needs for globally connected businesses and the World Wide Web. 10Gb/s Ethernet is now being adopted to meet increasing bandwidth needs for in-datacenter communications. However, most datacenter network architectures are still based on specialized hierarchical edge-core topologies which are costly to build, difficult to maintain and consume large amounts of power. This paper describes enhancements to Layer two Ethernet switches that support multipath L2 routing for scalable datacenters. This enables a cost-effective scalable network architecture based on enhanced layer two Ethernet switches. The architecture provides multipath routing for Ethernet while preserving important Ethernet features and interoperability with traditional Ethernet gear. While this work is currently limited to scalable fat tree networks, these topologies provide an attractive approach for scaling large datacenter networks.

    Read the paper…

    Comments are welcome: feedback@fulcrummicro.com


    ‘Fat tree’ switch architectures trim data center costs

    April 20th, 2009

    Researchers at University of California, San Diego determined that the use of commodity, high-performance Ethernet switches in a fat-tree configuration improves performance and saves money vs. the traditional multi-tier approach to networking.

    Instead of investing in specialty gear, such as 10 Gigabit Ethernet switches and routers, and using a standard three-tier architecture, Vahdat says companies could use commodity Ethernet switches in a “fat-tree” architecture at a much lower cost to achieve the same performance. Anecdotally he explains a 20,000 node network using pricey switches could cost as much as $28 million to construct; with commodity gear, the same network would ring up at closer to $4 million.

    As the leading enabler of 10GbE fat-tree switch architectures, we couldn’t agree more.  Read the article…

    Comments are welcome: feedback@fulcrummicro.com


    Fulcrum joins Cavium Octeon II processor launch

    April 14th, 2009

    Cavium took the wraps off its much-anticipated Octeon II Multicore MIPS64 Processor Family in unique fashion, with a partner-supported product launch.  With several customers in common and a new processor family that perfectly complements our FocalPoint Ethernet switches, Fulcrum was pleased to participate.  Rajiv Khemani, VP and GM of Cavium’s Networking and Communications Division commented about the partnership:

    Cavium is pleased to be partnering with Fulcrum to enable next-generation networking platforms.  The high performance of Fulcrum’s FocalPoint lossless Ethernet fabrics combined with the high programmability provided by the Cavium OCTEON II Processors enables networking equipment providers to deliver differentiated and extensible platforms that are suitable for the most demanding data center, cloud networking, and broadband access applications.

    Read the press release…

    Comments are welcome: feedback@fulcrummicro.com


    Lossless Ethernet Enables Telecom Backplanes

    April 1st, 2009

    Telecom backplanes have long been the territory of proprietary switch fabric solutions. Ethernet, with it’s lack of high bandwidth ports along with it’s lack of congestion management and flow control mechanisms was previously deemed unfit for these applications. All of this has changed over the last several years.

    During the Telecom bubble, there were over 20 proprietary switch fabric solutions on the market. This has been reduced to one or two. There is still a need for new, high bandwidth backplane fabrics, so why has there not been a resurgence of proprietary switch fabric chip sets? One major reason is the emergence of Lossless 10GbE Ethernet as a Telecom backplane solution. Products like the FocalPoint FM3000 series, provide advanced congestion management, scheduling and flow control that is required for these applications, at a price point that proprietary solutions can’t touch.

    Another indication of this industry change is the number of CPU and NPU vendors that have added 10GbE interfaces to their silicon. Even the last remaining proprietary fabric vendors have seen the writing on the wall and are adding Ethernet ports to their products.

    Comments are welcome: feedback@fulcrummicro.com