The Experts below are selected from a list of 6 Experts worldwide ranked by ideXlab platform

Karthik Ramasamy - One of the best experts on this subject based on the ideXlab platform.

  • Chapter 16 – Switch Fabric
    Network Routing, 2018
    Co-Authors: Deep Medhi, Karthik Ramasamy
    Abstract:

    A switch fabric is a core component of any router that provides a physical path between the ingress Linecard and the Egress Linecard. In this chapter, we present different types of backplanes that facilitate the movement of packets from one Linecard to another Linecard in a router. At a very high level, such switches can be broadly categorized as shared backplanes or switched backplanes. In a switched backplane, we consider single stage fabrics – shared memory and crossbar. We cover the scheduling algorithms for crossbar such as take-a-ticket, PIM, and iSLIP and analyze their pros and cons. We then discuss multistage switching fabrics, CLOS and Benes, and the complexity of scheduling algorithms in these switches. Finally, we explain various techniques about how to scale switches for higher link speeds.

Deep Medhi - One of the best experts on this subject based on the ideXlab platform.

  • Chapter 16 – Switch Fabric
    Network Routing, 2018
    Co-Authors: Deep Medhi, Karthik Ramasamy
    Abstract:

    A switch fabric is a core component of any router that provides a physical path between the ingress Linecard and the Egress Linecard. In this chapter, we present different types of backplanes that facilitate the movement of packets from one Linecard to another Linecard in a router. At a very high level, such switches can be broadly categorized as shared backplanes or switched backplanes. In a switched backplane, we consider single stage fabrics – shared memory and crossbar. We cover the scheduling algorithms for crossbar such as take-a-ticket, PIM, and iSLIP and analyze their pros and cons. We then discuss multistage switching fabrics, CLOS and Benes, and the complexity of scheduling algorithms in these switches. Finally, we explain various techniques about how to scale switches for higher link speeds.

Nick Mckeown - One of the best experts on this subject based on the ideXlab platform.

  • Optics inside Routers
    2003
    Co-Authors: Nick Mckeown
    Abstract:

    this paper we describe some recent developments in how optics can be used inside Internet routers to scale capacity and reduce power. At Stanford University we are currently designing a 100Tb/s Internet router with an optical switch fabric that guarantees 100 % throughput for all traffic. Background It has been speculated for some time that electronic packet-switched Internet routers will eventually be replaced by all-optical routers that process, buffer, route, and switch packets in the optical domain. There are a number of reasons why this won't happen anytime soon. First and foremost, an Internet router requires large random access buffers to hold packets during times of congestion. A widely used rule-ofthumb is that a router needs enough buffers to hold the data sent in an average round-trip-time (100-300ms). 1 When added to all the other features a router performs, a typical 10Gb/s Linecard has 30 million gates, 300 Mbytes of packet buffers, stores a million addresses, and consumes 200W. Unless features are radically pruned, optical Linecards won't be feasible any time soon. But optical components are already being used inside routers- to interconnect a router’s subsystems. To reduce power-density, there is a trend towards multirack routers [3,2,5], with typically 16-32 Linecards sharing a rack, and connected together by multimode parallel optical links and a central switch fabric. This has the advantage of reducing power density by spreading the router over multiple racks, but the total power is increased because of the increased number of conversions between the optical and electrical domains. The typical processing sequence is that arriving packets are processed by the ingress Linecard, and buffered until a central scheduler grants them access to the switch fabric; at which point they are converted into the optical domain, transmitted to the switch fabric, converted back into the electrical domain, switched by electronic crossbars, converted into the optical domain, transmitted to the Egress Linecard, converted back into the electrical domain, and then processed and scheduled for departure. An obvious question to ask is: Can the number of conversions- and the total system power- be reduced by replacing the electronic switch fabric wit