Overview:
The QFX10000 line of modular data center spine and core Ethernet switches delivers industry-leading scale, flexibility and openness, with a design that enables the seamless transition from 10GbE and 40GbE interface speeds to 100GbE and beyond. These high-performance, forward-looking switches are designed to help cloud and data center operators extract maximum value and intelligence from their network infrastructure well into the future.
Product Description
The Juniper Networks QFX10000 line of modular Ethernet switches delivers up to 96 Tbps of system throughput, scalable to over 200 Tbps in the future, to meet the rapid and ongoing traffic growth in the data center. Industry-leading scale and density on the QFX10000 modular switches redefine per-slot economics, enabling customers to do more with less while simplifying network design and reducing OpEx. Based on purpose-built Q5 ASICs, the QFX10000 line delivers unparalleled intelligence and analytics, providing deeper insights into application performance.
The QFX10000 line can be deployed in a number of different network designs and fabrics, including Layer 3 fabric and Juniper MC-LAG for Layer 2 and Layer 3 networks, giving customers complete architectural flexibility. Additionally, the open architecture ensures that customers can innovate on top of Juniper Networks Junos operating system to accelerate the pace of innovation.
The QFX10000 is also a key component of Juniper Networks MetaFabric Architecture, which provides a simple, open and smart approach to building data center networks. Additionally, the QFX10000 supports Data Center Interconnect (DCI), critical to workload mobility and application availability.
For networks evolving to SDN, the QFX10000 can integrate with VMware NSX SDN controllers and can act as a Virtual Extensible LAN (VXLAN) Layer 2 and Layer 3 gateway. The open and standards-based QFX10000 switches can also interoperate with the Open vSwitch Database (OVSDB) to support automated management and control capabilities. The QFX10000 line can integrate with OpenContrail SDN controller to allow users to choose their preferred SDN systems.
Two QFX10000 modular chassis options are available, providing full deployment flexibility:
- QFX10008 Ethernet Switch, an 8-slot, 13 U chassis that supports up to eight line cards
- QFX10016 Ethernet Switch, a 16-slot, 21 U chassis that supports up to 16 line cards
Both QFX10000 modular chassis can accommodate any combination of the following QFX10000 Ethernet line cards:
- QFX10000-36Q, a 36-port 40GbE quad small form-factor pluggable plus transceiver (QSFP+) or 12-port 100GbE QSFP28 line card
- QFX10000-30C, a 30-port 100GbE QSFP28/40GbE QSFP+ line card
- QFX10000-60S-6Q, a 60-port 1GbE/10GbE SFP/SFP+ line card with six-port 40GbE QSFP+ / two-port 100GbE QSFP28
Fully configured, a single QFX10016 chassis can support up to 480 100GbE ports, delivering the industry’s highest line-rate 100GbE port densities in its class. The QFX10000 switch fabric is capable of delivering up to 7.2 Tbps (full duplex) per slot, and the midplane-less orthogonal interconnect architecture ensures system longevity.
QFX10000 Modular Switches Highlights
- Industry-leading line-rate 100GbE port density with up to 480 100GbE ports in a single chassis
- Up to 96 Tbps Layer 2 and Layer 3 performance, scalable to over 200 Tbps in the future
- Unparalleled investment protection with high density 10GbE, 40GbE, and 100GbE; 400GbE ready
- System longevity with midplane-less orthogonal interconnect architecture
- Highest logical Layer 2 / Layer 3 scale; up to 1M MACs, 2 million host routes, 2 million FIB
- Deep buffers with up to 100ms packet buffering per port
- No head-of-line blocking with virtual output Queue (VoQ)- based architecture
- Flexible network architectures including Layer 3 fabric and Juniper MC-LAG for Layer 2 and Layer 3 networks
- Juniper Virtualized Open Network Operating System framework for programmability through APIs
- High availability with non-stop routing (NSR), non-stop bridging (NSB), and graceful route engine switchover (GRES)
- Rich automation capabilities with operations and event scripts, Python, Chef, and Puppet
Architecture and Key Components:
The QFX10000 modular data center spine and core Ethernet switches share a number of architectural elements. The Control Boards (CBs) employed by these switches run Junos OS, which processes all Layer 2 and Layer 3 protocols, while the Switch Fabric modules manage the chassis and provide switching functionality for data traffic coming from line cards.
The QFX10000 line cards, which are common across all modular QFX10000 platforms, include Packet Forwarding Engines (PFEs) that process network traffic, as well as a line-card processor that provides scalable local control.
The QFX10000 Virtual Output Queue (VOQ)-based architecture is designed to scale to very large deployments, with no headof-line blocking, a single-tier low-latency switch fabric, efficient multicast replication handling, and deep buffering to ensure performance at scale. The horizontal line cards in the front of the chassis directly connect with the vertical switch fabric cards in the rear of the chassis via orthogonal interconnects, without the need for a midplane. This midplane-less architecture ensures a smooth upgrade process to higher-speed switch fabric cards in the future, providing unparalleled investment protection. The system is designed from the ground up to support 400GbE in future.
To maintain uninterrupted operation, the QFX10000 fan trays cool the line cards, Control Board, and Switch Fabric modules with redundant, variable-speed fans. In addition, the QFX10000 power supplies convert building power to the internal voltage required by the system.
All QFX10000 components are hot-swappable, and all central functions are available in redundant configurations, providing high operational availability by allowing continuous system operation during maintenance or repairs.
QFX10000 Line Cards
The QFX10000 line cards support an extensive set of Layer 2 and Layer 3 services that can be deployed in any combination of L2- L3 applications.
Each QFX10000 line card is unique in its ability to support tri-speed 10GbE, 40GbE and 100GbE connections, enabling customers to transition seamlessly from one speed to another as their needs change. The unprecedented port density, high logical table sizes and up to 100ms deep packet buffers allow customers to deploy the most scalable systems necessary to meet the exponential data growth in the data center.
Each QFX10000 line card is built upon the Juniper Q5 silicon, which supports a wide range of Layer 2 and Layer 3 Ethernet functionality including 802.1Q VLAN, VXLAN, link aggregation, Virtual Router Redundancy Protocol (VRRP), L2 to L3 mapping, and port monitoring. Additionally, the line cards support filtering, sampling, load balancing, rate limiting, class of service (CoS), MPLS, Fibre Channel over Ethernet (FCoE) transit functionality, and other key features needed to deploy a dependable, lossless, high-performance Ethernet infrastructure.
QFX10000 Control Board
The QFX10000 Control Board supports control and management plane functionality with an integrated Routing Engine (RE) that features a quad-core, 2.5 GHz Intel processor with 32 gigabytes of SDRAM and an on-board solid-state drive (SSD) providing 32 GB of storage for Junos OS images and logs. The CB also has a front SSD slot to support a hot swappable SSD for secondary images and other external storage. The CB features AUX, console, and Ethernet ports on the front panel to support out-of-band system management and monitoring, while an external USB port provides a removable media interface to install Junos OS images manually.
The CB’s central CPU performs all system control functions and maintains hardware forwarding table and routing protocol states for the QFX10000 modular switches. Dedicated hardware on the CB module supports chassis management functions such as environmental monitoring, while communication between CB modules and individual line cards takes place over a dedicated internal 10GbE out-of-band control interface.
QFX10000 Switch Fabric
The QFX10000 Switch Fabric modules are hot-swappable and serve as the central non-blocking matrix through which all network data passes.
Both the QFX10008 and QFX10016 systems have six Switch Fabric cards that provide N+1 redundancy. The Switch Fabric cards deliver 7.2 Tbps throughput per slot. The switch cards can be upgraded to provide more than 200 Tbps of system throughput in the future.
Switch Fabric modules perform the following key functions:
- Monitor and control system functions
- Interconnect all line cards
- Handle clock and system resets
Power
The Premium and Advanced software licenses are classified as Class 1, Class 2, Class 3, etc., and offered as perpetual licenses. Class 1 licenses are applicable to the QFX5110-48S and QFX5110-32Q switches. Please see Ordering Information for license SKU descriptions. The AC supplies on the QFX10008 chassis accept 200 to 240 volts alternating current (VAC) input and deliver 2,700 watts of power to the chassis, while the DC power supplies accept -40 to -72 volts direct current (VDC) input and deliver 2,500 watts of power to the chassis. Each AC and DC power supply has two inputs for feed redundancy.
Cooling
The QFX10000 modular chassis support front-to-back cooling with air being pulled in through the perforations on the Control Boards and the line cards in the front of the chassis. Hot air is exhausted through the fan trays that are placed in front of the fabric cards in the rear of the chassis as well as through the power supplies. Each chassis hosts two fan tray subsystems with redundant fans. Each fan tray subsystem consists of a fan tray controller and a fan tray. The fan tray controller provides power to the fan trays and manages the fans in the fan tray.
Flexible Network Architectures
The QFX10000 switches support a diverse set of deployment options, including Layer 3 fabric and Layer 2 and Layer 3 networks with Juniper multichassis link aggregation group (MC-LAG). Customers can choose the architecture that best suits their deployment needs and easily adapt and evolve as requirements change over time. The QFX10000 switches serve as the universal building block for all of these switching architectures, enabling data center operators to build cloud networks their way.
- Layer 3 Fabric: For customers looking to build scale-out data centers, a Layer 3 spine and leaf Clos Fabric is ideal due to its non-blocking and predictable performance and scale characteristics. As an example, a two-tier fabric with the QFX10000 switches as the spine and QFX5100 switches as leafs can scale to support more than 50,000 10GbE server ports at 3:1 oversubscription. One of the most complicated tasks in building an IP fabric is assigning all of the details such as IP addresses, BGP AS numbers, routing policy, loopback addresses, and many other implementation details. Automating the creation of an IP fabric at a large scale is also difficult. Juniper has created the OpenClos project to provide free, open-source tools to automate the creation of IP fabrics in the data center. OpenClos is a set of Python scripts developed as an open-source project that live on GitHub. OpenClos takes a set of inputs that describe the shape and size of a data center and outputs network switch configuration files and even a cabling plan.
- Juniper MC-LAG: The QFX10000 switches also eliminate Spanning Tree Protocol in traditional Layer 2 networks when deployed in a MC-LAG configuration. The active-active operation of MC-LAG ensures complete bandwidth utilization between the network’s access and aggregation layers, while the dual control plane technology ensures highest availability for applications.
- Overlays: Customers can deploy overlay networks to provide Layer 2 adjacencies for applications over Layer 3 fabrics. The overlay networks use VXLAN in the data plane and EVPN or OVSDB for programming the overlays. The overlays can operate without a controller, or can be orchestrated with an SDN controller like OpenContrail or VMware NSX.
Carrier-Class Operating System
The QFX10000 line of switches runs the same Junos OS used by all other Juniper Networks EX Series and QFX Series Ethernet Switches, as well as the Juniper Networks routers that power the world’s largest and most complex networks.
By using a common operating system, Juniper delivers a consistent implementation and operation of control plane features across all products. Junos OS employs a highly available modular architecture that prevents isolated failures from bringing down an entire system. Key Junos OS features that enhance the functionality and capabilities of the QXF10000 include:
- Software modularity, with process modules running independently in their own protected memory space and with the ability to do process restarts • Uninterrupted routing and forwarding with features such as nonstop routing (NSR) and nonstop bridging (NSB)
- Commit and rollback functionality that ensures error-free network configurations
- A powerful set of scripts for on-box problem detection, reporting, and resolution
Figure 1: QFX10000 modular switches can be deployed in Layer 3 fabric or MC-LAG configurations.
Juniper Virtualized Open Network Operating System Framework
The QFX10000 switches have an open software architecture that enables customers to innovate along with Juniper to accelerate the pace of innovation. Customers can create and run applications alongside Junos; these applications can then follow their own software release schedule.
This new solution offers modularity and direct programmability through APIs. The platform and Packet Forwarding Engine (PFE) modules are separate from the control module(s). Customers can create and run guest applications, either as a VM or a Linux daemon, alongside Junos and communicate with Junos and also directly with platform and PFE modules through normalized programmable APIs. Having the ability to build applications that can program the QFX10000 control plane, data plane and the platform itself, the customers can implement specific and unique network forwarding functions and automate the control and management by integrating with the customer’s SDN controller and orchestration tools. The Juniper QFX10000 programmable APIs have a stable abstraction so that the customer applications can work across Juniper platforms as well as future Junos releases. Based on Apache Thrift, the Juniper QFX10000 supports standardized API definition via Interface Description Language (IDL) while allowing the use of customer’s favorite programming language.
Other services such as analytics and automation can also run as separate daemons and can be directly accessed through orchestration tools.
Figure 2: Juniper Virtualized Open Network Operating System Framework
Junos Space Network Director
The QFX10000 switches can be managed through Junos Space Network Director, a next-generation network management solution that allows users to visualize, analyze, and control the entire enterprise network—data center and campus, physical and virtual—through a single pane of glass. Network Director incorporates sophisticated analytics for real-time intelligence, trended monitoring, and automation to increase agility as well as faster rollout and activation of services.
For cloud deployments, Network Director provides a set of REST APIs that enable on-demand and dynamic network services by simplifying the consumption of services for multitenant environments. With third-party cloud orchestration tool integration, the Network Director API enables automation and provisioning of Layer 2, Layer 3, and security services in the data center without the need for manual operator intervention.
Features and Benefits:
High Availability
QFX10000 modular spine and core switches deliver a number of high availability features that ensure uninterrupted, carrier-class performance. Each QFX10000 chassis includes an extra slot to accommodate a redundant RE module that serves as a backup in hot-standby mode, ready to take over in the event of a master RE failure. If the master fails, the integrated Layer 2 and Layer 3 graceful Routing Engine switchover (GRES) feature of Junos OS, working in conjunction with the nonstop active routing (NSR) and nonstop bridging (NSB) features, ensures a seamless transfer of control to the backup, maintaining uninterrupted access to applications, services, and IP communications.
Virtual Output Queue (VOQ)
The QFX10000 switches support a virtual output queue (VOQ)- based architecture designed for very large deployments. VOQ refers to a queue on the egress port that is maintained by the ingress PFE. With VOQ architecture, packets are queued and dropped on ingress during congestion with no head-of-line blocking.
Automation
The QFX10000 switches support a number of features for network automation and plug-and-play operations. Features include operations and event scripts, automatic rollback, and Python scripting. The switches can also support integration with VMware NSX, OpenContrail, Puppet, OpenStack, and CloudStack.
MPLS
QFX10000 switches support a broad set of MPLS features, including L3 VPN, IPv6 provider edge router (6PE, 6VPE), RSVP traffic engineering, and LDP to allow standards-based network segmentation and virtualization.
VXLAN
The QFX10000 supports Layer 2 and Layer 3 gateway services that enable VXLAN-to-VLAN connectivity at any tier of the data center network, from server access to the edge. The QFX10000 can integrate with NSX through data plane (VXLAN) and control and management plane (OVSDB) protocols to centrally automate and orchestrate the data center network.
FCoE
As Fibre Channel over Ethernet (FCoE) transit switches, the QFX10000 line provides an IEEE data center bridging (DCB) converged network between FCoE-enabled servers and an FCoE-enabled Fibre Channel storage area network (SAN). The QFX10000 offers a full-featured DCB implementation that provides strong monitoring capabilities, helping SAN and LAN administration teams maintain clear management separation. The FCoE transit switch functionality, including priority-based flow control (PFC) and Data Center Bridging Capability Exchange (DCBX), are included as part of the default software.
