Extreme Networks: Building Cloud-Scale Networks Using Open Fabric Architectures A SOLUTION WHITE PAPER

Extreme Networks: Building Cloud-Scale Networks Using Open Fabric Architectures A SOLUTION WHITE PAPER

WHITE PAPER Building Cloud- Scale Networks Abstract TABLE OF CONTENTS Introduction 2 Open Fabric-Based Approach to Data Center Networks 3 Industry Open Initiatives 5 Conclusion 6 Several technology inflection points are coming together that are fundamentally changing the way networks are architected, deployed and operated both in the public cloud and private cloud. Requirements are rapidly changing and driving new approaches to building data center networks. Extreme Networks is enabling next generation data centers with an open, standards-based, high-speed, interconnectivity fabric architecture. This white paper outlines Extreme Networks Open Fabric approach to data center networking including high speed connectivity, low latency, multi-path mesh-type connectivity, high resiliency and support for network and storage convergence. The paper goes on to explain ways to evolve your data center network with virtualization and automation and how to harness the benefits of lower power consumption and other features of the Extreme Networks architecture toward a greener data center. Introduction Several technology inflection points are coming together that are fundamentally changing the way networks are architected, deployed and operated both in the public cloud as well as the private cloud. From performance, to scale, to virtualization support and automation to simplified orchestration, the requirements are rapidly changing and driving new approaches to building data center networks. On one hand the need for network performance is ever increasing. With computational density intensifying, the number of virtual machines (VMs) per server is growing fast. Where earlier 4-8 VMs per server were common, today cloud providers are moving to 16, 32 or more VMs per server. This is driving increased bandwidth demand directly at the server-network edge. As this bandwidth demand at the server-network edge grows, 10 GbE is expected to become the preferred connectivity speed there, with the aggregation or core moving to 40 GbE. Additionally, with more enterprises looking to the cloud to add capacity on demand, traffic patterns within the cloud are assuming more east-west characteristics with VM-VM traffic within a server, within a rack, and across racks, all increasing. This increasing east-west traffic pattern mix is driving increased requirements for low latency and high speed connectivity within the cloud. Finally, the confluence of network and storage traffic on a common Ethernet fabric is driving the need for more predictable network performance in order to ensure that storage performance is not compromised due to the transient burstiness inherent in data traffic patterns. Cloud-Scale Networks White Paper 2

All of the above factors are driving toward higher performance, lower latency, and more predictable network architectures for both private and public clouds. In parallel, the move to a more dynamic data center is well underway where VMs can be moved at will within and across data centers (indeed at times even without user intervention based on criteria such as capacity and utilization) and users can dynamically add and reduce capacity on-demand. This type of dynamism is driving a strong requirement for the network infrastructure to scale on demand, provide the capability for customization and automation, as well as integration with server virtualization technologies to reduce manual provisioning and configuration. The notion of orchestrating the data center where server, network and storage resources can all be provisioned in an automated manner is gaining momentum. An Open Fabric-Based Approach to Data Center Networks Given the changing landscape and evolving requirements, data center architectures are moving toward a fabric-based approach to providing high speed, low latency connectivity within the data center. Several attributes typically define a fabric -type network architecture: High-speed connectivity Low latency from port to port Multi-path mesh-type connectivity High resiliency However, in today s cloud environments there are several additional attributes that become part of the network fabric. These include: Support for network and storage convergence Support for virtualization, automation and customization Ease of network provisioning, configuration and management Low power One of the key attributes of cloud providers is that infrastructure is their cost of goods sold. As such, the ability to maintain pricing leverage remains a key driver to cloud providers ability to reduce the cost of their infrastructure. This leads toward an open and interoperable approach to building network infrastructure rather than a vendor-specific proprietary architecture. Furthermore, with technology changing rapidly in the data center, the lock-in of proprietary vendor-specific technology can be very costly as technology directions shift rapidly in this dynamic landscape. The industry is rapidly moving toward open architectures for the cloud with several industry consortiums paving the way by providing reference models, technologies and tools for building large, scale-out architectures. Indeed, the technology components for an open fabric for the cloud are already in place. Some of these components are described below. OPEN, STANDARDS-BASED, HIGH-SPEED, INTER- CONNECTIVITY FABRIC From a fabric interconnectivity perspective, standards-based 10 GbE and 40 GbE interconnectivity fabrics are fast becoming the mainstay of the data center network. With the server edge moving to 10GbE, the aggregation layer is moving to 40 GbE. This requires high density 10 GbE as well as high density, high-performance 40 GbE connectivity solutions. Along with density and capacity, low latency and low carbon footprint are becoming key requirements. Extreme Networks BlackDiamond X series chassis will offer up to 768 wire speed 10 GbE ports or up to 192 wire speed 40 GbE ports in just a 1/3 rack size form factor. This level of non-blocking performance and density is industry leading and the basis for building cloud-scale connectivity fabrics. Using this model, servers can directly attach to a high density 10 GbE End-of-Row (EoR) solution (such as the BlackDiamond X series), or may connect to a tier of Top-of- Rack (ToR) switches (such as Extreme Networks Summit X670) with the TOR switches then connecting over multiple 40 GbE links to the aggregation or core layer. M-LAG 40G M-LAG 40G LAG or NIC Teaming Up to 768 10 GbE Servers Up to 128,000 Virtual Machines LAG NIC 10G Teaming Up to 4,560 10 GbE Servers Up to 128,000 Virtual Machines 10G 5624-01 5625-01 FIgure 1 FIgure 2 Cloud-Scale Networks White Paper 3

Most of the open, standards-based solutions coming on the market today for high-speed interconnectivity also provide low latency cut-through switching capabilities. For example the Summit X670 ToR switch offers latency of around 800-900 µsec while the BlackDiamond X chassis will offer port-to-port latency of well below 3 µsec. This combination allows building single tier or two-tier network fabrics that offer very low end-to-end latency. To address resiliency, where dual uplinks are used from the TOR to the aggregation tier, solutions such as M-LAG may be used for active-active redundancy. Similarly if servers need to be dual homed to the ToR or EoR tier, NIC teaming can be used in combination with an M-LAG-type approach for activeactive redundancy. While M-LAG itself is proprietary, the tier that dual-homes into the M-LAG switches simply uses standard link aggregation. For example, servers can use link aggregation (or NIC teaming as it is commonly called) to dual-home into two TOR switches which present themselves as a single switch to servers via M-LAG. (Reference: Exploring New Data Center Network Architectures with Multi-Switch Link Aggregation (M-LAG)). If a true multi-homed architecture is to be used, for example where 4 uplinks may connect to 4 different switches, a standards track protocol such as TRILL (Transparent Interconnection of Lots of Links) may be used to provide Layer 2-based multipath capability. However, with data centers typically dual-homing connections at each layer, an M-LAG-type approach should suffice. In addition to high-speed, low-latency, and high-density fabric connectivity, many of the interoperable standards-based solutions on the market today are also low carbon footprint switching infrastructures. This is particularly important as the cloud network transitions from a 1 GbE edge to a 10 GbE edge and 40 GbE core, since the power footprint of the network increases significantly with this transition. Where earlier 10 GbE ports would consume 10W-30W per port, today that number is dropping rapidly to around 3W-10W per port. For example, the BlackDiamond X chassis will consume around 5W per 10 GbE port. OPEN APPROACH TO VIRTUAL MACHINE SWITCHING The broad adoption of server virtualization has been instrumental in enabling the cloud model to gain acceptance. However, along with the benefits of virtualization come a set of challenges. For example addressing VM switching through virtual switches gives rise to several challenges. From the complexity of dealing with multiple hypervisor technologies, to providing security between VMs within the hypervisor, to software (CPU)-based switching between VMs that could lead to unpredictable performance, the list of potential issues runs large. The IEEE 802.1Qbg working group is looking at this problem and is defining new forwarding modes that allow switching VM traffic directly in the network switch. (Reference: VEPA: An Answer to Virtual Switching). The ability to support this new forwarding mode in the network infrastructure provides an open standards track approach to simplifying VM switching. For example, data center networking products from Extreme Networks support the ability to switch VMs in hardware at wire speed. The Summit X670 as well as the BlackDiamond X series products will be able to switch up to 128k VMs in hardware at wire speed. The ability to leverage standards track technology provides investment protection without the lock-in associated with proprietary, vendor-specific technologies. In addition, the ability to address VM configuration and mobility in a hypervisor agnostic manner is also becoming important. The IEEE 802.1Qbg working group is looking at extensions that address some of the network mobility challenges around VMs through the concept of defining network profiles associated with VMs which can then be moved as a VM moves. Extreme Networks XNV technology supports the notion of Virtual Port Profiles (VPPs) for VMs which can then be applied to VMs on any hypervisor. This provides a hypervisor-agnostic way of provisioning network characteristics for VMs. In addition, XNV can also automatically migrate VPPs and enforce them on any target switch to which the VM moves. OPEN APPROACH TO CONVERGENCE IN THE DATA CENTER Much interest has been focused around network and storage convergence in the data center. Storage and network convergence is becoming a reality today using 10 G or higher speed Ethernet in conjunction with iscsi-based storage which works natively over an Ethernet based TCP/IP infrastructure. FCoE- based storage convergence is a little further out in terms of its adoption and interoperability. However, in both cases, the availability of standards-based Data Center Bridging (DCB) technology is a key facilitator to enabling this convergence. DCB allows partitioning of traffic into multiple traffic classes on a common Ethernet fabric and then assigns priority to those traffic classes, as well as specifying bandwidth parameters to traffic classes. In essence, using open, standards-based DCB technology, storage traffic can be merged on to a common data Ethernet LAN while maintaining a degree of separation between the two. Indeed, not just storage traffic but other classes of traffic may also run on a common Ethernet fabric while providing isolation between the various traffic classes. As an example, management traffic or Vmotion traffic may also be moved to a common converged Ethernet fabric and leverage DCB capabilities to ensure reliable and predictive traffic behavior. As such, leveraging DCB standards-based technology provides a solid foundation for network and storage convergence that allows scaling deployments in the cloud while continuing to provide the predictability, performance, and latency required to meet storage performance. This technology is available today from several vendors, across switches, initiators and targets. Extreme Networks has been shipping DCB-enabled infrastructure for some time now and has participated in various industry interoperability forums focused around DCB. Cloud-Scale Networks White Paper 4

LEVERAGING OPENFLOW TO SIMPLIFY CLOUD-SCALE PROVISIONING OpenFlow is a relatively new industry-backed technology that centralizes the intelligence in the network while keeping the data path distributed. By centralizing the intelligence, OpenFlow provides a platform upon which a diverse set of applications can be built and used to program, provision and manage the network in a myriad of different ways. Within the context of a data center fabric, OpenFlow holds the promise of taking complex functions such as traffic provisioning in converged data center networks, logical network partitioning in public and hybrid cloud environments, as well as user and VM provisioning in highly virtualized data centers, providing a centralized and simplified approach to addressing all of these at scale. OpenFlow is in its early stages in terms of applications and adoption. It will take time for the technology to mature. But, it holds a lot of promise as a platform upon which smart applications can be built for the next generation data center fabric. In effect, OpenFlow provides an open source platform upon which users can customize, automate, and innovate to provide new ways to address some of the challenges in the data center. It is important to note that the benefits of OpenFlow are not limited to data center and cloud scale networks. Indeed, applications are being built on the enterprise and campus side of the network as well using OpenFlow technology. But within the context of the data center and the cloud infrastructure, OpenFlow holds particular promise for both simplifying and automating complex provisioning tasks and easing the burden on network administrators. There are two pieces to the OpenFlow solution. The first is the availability of switching and fabric infrastructure that support the OpenFlow protocol to provision flow entries in the fabric. The second is the OpenFlow Controller which controls and programs the OpenFlow switching infrastructure. OpenFlow controllers are being made available through open source initiatives, as well as through commercial vendors who are building specific solutions for problems using their own OpenFlow controllers. Extreme Networks is implementing OpenFlow technology in its data center switching infrastructure and working with a variety of different OpenFlow controllers to bring unique and differentiated solutions to market. Industry Open Initiatives OPEN NETWORKING FOUNDATION The Open Networking Foundation (ONF) is a newly formed organization that is focused on technology for software-defined networking, of which OpenFlow is a key technology. The member list of the ONF is a broad spectrum of both the producers i.e. vendors, as well as consumers of networking technology. Software defined networking and OpenFlow hold the promise of revolutionizing the networking industry with an open-source, industry-backed platform upon which innovative solutions that address key industry problems can be built. Extreme Networks is a member of ONF. Extreme Networks currently plans to implement OpenFlow technology in its operating system ExtremeXOS for its data center line of switches and also participated in a live OpenFlow demonstration at the 2011 Las Vegas Interop Conference. Extreme Networks will work with a variety of OpenFlow controllers from open source controllers to commercial controllers to bring to market solutions that address a variety of needs such as virtual network provisioning for cloud customers. OPENSTACK OpenStack is an open source software development community that delivers a very scalable cloud operating system. The OpenStack community has three parallel tracks: OpenStack Compute: OpenStack Compute is open source software designed to provision and manage large networks of VMs, creating a redundant and scalable cloud computing platform. It provides the software, control panels, and APIs required to orchestrate a cloud, including running instances, managing networks, and controlling access through users and projects. OpenStack Compute strives to be both hardware and hypervisor agnostic, currently supporting a variety of standard hardware configurations and the major hypervisors. OpenStack Object Storage: OpenStack Object Storage is open source software for creating redundant, scalable object storage using clusters of standardized servers to store petabytes of accessible data. OpenStack Image Service: OpenStack Image Service provides discovery, registration, and delivery services for virtual disk images. With so many different technology inflection points coming together in parallel, the acceptance and adoption of open, interoperable technologies in the cloud is gaining momentum. Not just within the networking space, but across servers, networking and storage. While there have been several industry driven open initiatives, two of them stand out as getting significant mindshare. Cloud-Scale Networks White Paper 5

Extreme Networks is part of the OpenStack community focusing on bringing solutions to network provisioning in large cloud-scale networks. The challenge with compute provisioning in large cloud-scale environments is that traditionally the network has been missing from the provisioning piece. However, the network fabric is increasingly becoming a key piece of the solution since a user s quality of experience depends largely on the network s ability to service that user, meet the user s SLAs, as well as provide isolation, protection and security between users. As a result, any solution to compute provisioning in the cloud now needs to include the network fabric. Extreme Networks is working with the OpenStack community toward an open approach to solving provisioning problems in the cloud infrastructure. Conclusion Several trends, from virtualization, to convergence, to power, are driving newer architectures and technologies in the data center. Very scalable, efficient, and high-performance fabric-based architectures can be built and deployed using open, interoperable and industry accepted approaches. From high-density 10 GbE to high-density 40 GbE, to support for virtualization and convergence, as well as redundancy and multipath capability, open industry accepted solutions are gaining a foothold in building cloud-scale architectures. Various industry consortiums are driving the acceptance of open and interoperable solutions. Extreme Networks participation in these industry consortiums and its commitment to furthering the cause of open networking technology finds further validation in its newly announced products such as the BlackDiamond X series and the Summit X670 which may help pave the way toward building open, fabric-based network architectures for the cloud infrastructure. http://www.extremenetworks.com/contact Phone +1-408-579-2800 2014 Extreme Networks, Inc. All rights reserved. Extreme Networks and the Extreme Networks logo are trademarks or registered trademarks of Extreme Networks, Inc. in the United States and/or other countries. All other names are the property of their respective owners. For additional information on Extreme Networks Trademarks please see http://www.extremenetworks.com/about-extreme/trademarks.aspx. Specifications and product availability are subject to change without notice. 1792-1213 WWW.EXTREMENETWORKS.COM Cloud-Scale Networks White Paper 6