Next-Generation Data Center Core Switches

Embed Size (px)

Citation preview

  • 7/28/2019 Next-Generation Data Center Core Switches

    1/9

    Next-generation Data Center Core Switches Public

    2012-11-26 HUAWEI Confidential Page 1, Total 9

    Next-generation Data Center Core Switches

    By Mark Wu

    Introduction

    Since the introduction of cloud computing technology in 2006, it has developed so

    rapidly that almost all enterprise IT services have migrated, or are in the process of

    migrating, to cloud-computing platforms.

    Data centers are seeing an annual growth rate of over 40% in network devices.

    Among all network devices in a data center, core switches are the key nodes in the

    architecture of the entire cloud-computing network.

    The reasons behind the development of Data Center Core Switches

    The development of data center core switches can be attributed to many factors. The

    primary driving force is the revolutionary transition from the client/server traffic

    model to the server/server traffic model. During this transition, Incast and multicast

    traffic models are more widely used on networks than previous unicast traffic models.

  • 7/28/2019 Next-Generation Data Center Core Switches

    2/9

    Next-generation Data Center Core Switches Public

    2012-11-26 HUAWEI Confidential Page 2, Total 9

    As more and more core enterprise services migrate to IT platforms, enterprises are

    increasing their spend on IT system construction. Development of data center

    technologies, such as large-scale server clusters, virtualization, and big data, also

    bring higher requirements for data center networks.

    Service

    (application)

    driving

    forces

    Cloud

    ComputingScalability

    Multi-service

    Integration

    High

    Reliability

    Application

    virtualization

    Fast expansion of

    data centers (scale up

    and scale out)

    Ethernet, SAN,

    and high-speed

    computing

    Core services

    on IT

    platforms,

    cluster

    systems

    Technical(vendor)

    driving

    forces

    Network

    adaptation to

    virtualization

    Innovation inhigh-performance

    hardware-platforms +

    control protocols

    Everything overEthernet

    (requiring10GE

    100GE access)

    ISSU, microkernel,

    distributed

    network

    The preceding table shows that the key to the success of data center switches is to

    combine service requirements with mature technologies. Huawei follows this

    principle when developing next-generation data center core switches.

    Characteristics of Current Data Center Core Switches

    Currently, the data center core switches of mainstream vendors have the following

    characteristics:

    1. High scalabilityUsage

    Scenario

    For rapid expansion of

    data centersFor VM migration

    For more flexible

    network architecture

    scaling

  • 7/28/2019 Next-Generation Data Center Core Switches

    3/9

    Next-generation Data Center Core Switches Public

    2012-11-26 HUAWEI Confidential Page 3, Total 9

    More accessible

    servers, higher

    accessibility, and

    higher throughput

    Scalable large L2

    network to support

    VM migration

    Modular scaling for

    data center networks,

    for easier network

    deployment,

    maintenance andmanagement

    Matching

    technology

    Higher-performance

    hardware

    New protocols and

    hardware

    components

    Pod-based network

    design

    Chips with higher

    processing

    capabilities, interfaces

    with higher speeds

    (40GE/100GE), andnetwork devices with

    higher port densities

    New L2 protocol

    TRILL (VLAN

    anywhere but not

    every where), and L2

    interconnection

    between data centers

    New data center

    network design

    fulfilling network

    requirements for scaleup and scale out

    2. Virtualization capabilities

    Usage

    Scenario

    To simplify network

    topology and facilitate

    network maintenance

    To implement

    resource sharing

    and flexible

    resource

    allocation

    To support migration

    of VMs and detection

    VMs

    Matching

    technology

    Two (multi)-node

    clusterVirtual switch

    802.1BR, 802.1Qbg,

    out-of-band NMS

    Well developed

    two-node cluster

    technology and

    two-node cluster

    solutions for

    mainstream vendors.

    Multi-node clustertechnology is more

    difficult and is still

    under development.

    Matrix expansion and

    virtual stacking

    technologies used on

    access switches can

    also establish clusters.

    Virtual switch

    technology helps

    faster network

    deployment and

    improves

    efficiency ofresource

    utilization,

    including

    equipment rooms,

    power supply

    systems, and line

    cards.

    All these technologies

    are currently under

    development, and

    major vendors have

    their own solutions. It

    is still uncertain

    which technology will

    become predominant.

  • 7/28/2019 Next-Generation Data Center Core Switches

    4/9

    Next-generation Data Center Core Switches Public

    2012-11-26 HUAWEI Confidential Page 4, Total 9

    3. Support for multi-service integration and network convergence

    UsageScenario

    For complex

    network services

    For integration of

    traditional servicesFor network convergence

    Multi-tenant,

    mobile IP, VPN,

    etc.

    Integration offirewall, network

    analysis, and load

    balancing services

    Convergence oftraditional FC network

    and HPC network with

    Ethernet networks

    Matching

    technology

    New protocols

    and chips (VPN

    technologies,

    VxLAN/NvGRE,

    IPv6, and so on)

    Integrated service

    line cards

    Everything Over

    Ethernet

    New chips have

    higher processing

    capabilities and can

    process complex

    services.

    Mainstream

    vendors have just

    started the

    development of

    new chips, and

    none of these chipsare widely used.

    Complex services

    cannot be integrated

    on ASIC. Therefore,

    these services are

    integrated into one

    single device by

    using integrated

    service line cards.

    New technologies, such

    as 10GE access and DCB,

    enable data from

    heterogeneous networks

    to be transmitted over

    Ethernet, for example,

    FCoE and RDMAoE.

    Deficiencies of Current Data Center Core Switches

    Although data center core switches have made many breakthroughs in services and

    technologies, they still have the following deficiencies:

    1. Insufficient network scalability

    At present, very few data center core switches can provide scalability thatsupports network expansion for the next five to 10 years. This is because the

    device architecture cannot keep pace with the rapid network expansion.

    Server virtualization also brings great demand in L2 data switching. However,L2 networks and nodes are not very scalable due to defects of the L2

    networks.

    Traditional L2 protocols, such as STP, only solve the problem of L2 loops anddo not determine how to establish a large L2 network. There are still no mature

  • 7/28/2019 Next-Generation Data Center Core Switches

    5/9

    Next-generation Data Center Core Switches Public

    2012-11-26 HUAWEI Confidential Page 5, Total 9

    solutions for L2 VM migration between data centers.

    2. Separate network virtualization and service virtualization

    Services and applications become more flexible and dynamic after servicevirtualization.

    It is a big challenge for virtual networks to dynamically change configurationsand deployments to adapt to changes in services and applications.

    3. Insufficient openness to network behavior

    As network application environments become more complex, many customerswant to customize their own network behaviors. However, customer networks

    have their own characteristics and requirements, and standard devices cannot

    meet the special requirements of all customers. To address this problem, some

    vendors have introduced the idea of using open standard interfaces to control

    network behavior.

    Openflow, OpenStack, and SDN technologies were developed following thisidea. They share network control details with external applications to support

    the customization of network behavior. Although these technologies still need

    improvement and have not been widely used on network devices, these

    technologies may be the way of the future.

    Other than technologies that separate service control from forwarding,customization of network behavior can also be implemented through APIs

    provided by network devices or through middleware provided by an open

    platform. The need for network behavior customization cannot be ignored

    regardless of the technology used.

    4. Duplicate investment on core switches of data centers and campus networks

    Separate core switches for data centers and campus networks ensure networksecurity. However, deploying two groups of core switches increases

    investment on physical devices and raises costs of device management and

    maintenance.

    For customers, using the same hardware platform for data center and campus

  • 7/28/2019 Next-Generation Data Center Core Switches

    6/9

    Next-generation Data Center Core Switches Public

    2012-11-26 HUAWEI Confidential Page 6, Total 9

    network services allows the implementation of uniform service management

    and the facilitation of network deployment, operation, and maintenance. The

    core of the data center and campus network are physically coupled through

    virtual switch technology.

    Technically, it is an inevitable trend for a data center and campus network toshare the same group of core switches.

    5. High electricity costs

    New core switches use more power than old core switches. Traditional powersupply systems in the data center equipment rooms cannot meet the

    requirements of these new switches. High power consumption results in high

    temperatures, which may degrade the reliability of the device.

    New switches use more power because they use complex, high-frequencychips to provide complex services.

    Therefore, it is important to reduce the power consumption of data center coreswitches. Energy-saving core switches improve device reliability, save energy,

    and have lower requirements for data center equipment rooms.

    Huaweis Next-generation Data Center Core Switches

    Huawei is dedicated to offering competitive data center core switches and data

    center network solutions for customers around the world. Huawei has developed the

    CloudEngine 12800 series, its next-generation data center core switches, which

    provide the following features to meet service requirements of customers in the

    cloud-computing era:

  • 7/28/2019 Next-Generation Data Center Core Switches

    7/9

    Next-generation Data Center Core Switches Public

    2012-11-26 HUAWEI Confidential Page 7, Total 9

    CE12812 CE12808 CE12804

    1. Scalable network: higher scalability

    By 2012, the largest switching capacity of a data center core switch will be480 Gbps per slot because the maximum capacity of a switch panel is 48 x

    10GE. If the expected service life of a core switch is five years, the switch

    must support a bandwidth upgrade to at least eight times the current

    bandwidth. That is, the switching capacity must increase to about 4 Tbps per

    slot in five years.

    Huawei CE12800 supports standard Layer 2 protocols, such as TRILL, usedto build large Layer 2 networks. In addition, the Layer 2 TRILL network can

    seamlessly connect to standard IP networks, allowing customers to flexibly

    deploy services on a large network.

    2. Virtual network: close coupling of application and network

    In-band protocols between server network adapters and network devices canquickly detect the migration of a large number of VMs, and network devices

    can also respond to VM migration. However, this is not enough for

    cloud-computing applications.

    To respond to changes in virtualized services, many devices, including loadbalancing devices and WAN routers, need to adjust their configurations. The

  • 7/28/2019 Next-Generation Data Center Core Switches

    8/9

    Next-generation Data Center Core Switches Public

    2012-11-26 HUAWEI Confidential Page 8, Total 9

    configuration adjustment may even involve multiple data centers.

    Therefore, network virtualization needs to be closely coupled with servicechanges using out-of-band network management interfaces. Inbound

    protocols can combine with out-of-band network management interfaces to

    implement end-to-end virtualization.

    3. Open network: comprehensive customization capabilities

    Huawei supports Openflow, OpenStack, and SDN technologies. Huawei also provides open service platforms with standard APIs or

    middleware to overcome defects in standard protocols. An open service

    platform allows customers to quickly deploy their own network services.

    4. Resource-sharing: one group of core switches for data centers and campus

    networks

    Huaweis CE12800 supports virtual switch technology, which can virtualize asingle switch into a maximum of eight local switches (the number will

    increase to 16 in the future). The excellent forwarding performance of the

    CE12800 will provide an integrated service platform supporting both data

    center services and campus network services.

    Physical coupling of core nodes helps customers realize uniformmanagement and deployment of services and a simplified operations and

    maintenance, which reduces a customers capital and operating expenditures.

    5. Environmentally-friendly design: build a green data center

    Huawei's next-generation data center switches are installed withenergy-saving ASIC chips and transceivers and offer the industry's most

    efficient digital power modules. Power consumption of key components can

    be adjusted based on traffic volume, which reduces power consumption per

    unit of transmission speed.

    Expanding the service life of network devices is an effective way to protectthe environment because abandoning a core switch would result in high

    carbon emissions, which is much more costly than the power fees of a core

  • 7/28/2019 Next-Generation Data Center Core Switches

    9/9

    Next-generation Data Center Core Switches Public

    2012-11-26 HUAWEI Confidential Page 9, Total 9

    switch. Huawei's next-generation data center switches support a capacity

    upgrade to at least eight times the current capacity. This scalability can

    support a service life of up to five to 10 years, saving resources and

    protecting customer investment.

    Huawei CE12800: Next-generation Data Center Core Switches

    Huaweis CE12800 series of are designed for cloud-computing data centers and

    super-large virtualized data center networks. The CE12800 has high bandwidth

    scalability and is capable of upgrading bandwidth from 1 Tbps to 4 Tbps per slot.

    Compared with other core switches, CE12800 is highly competitive in terms of its

    scalability, virtualization, openness, usability, and energy efficiency. The

    CE12800 series offer core switches for future-oriented data centers.