Supported Platforms
QFabric System Overview
The architecture of legacy data centers contrasts significantly with the revolutionary Juniper Networks data center solution.
This topic covers:
Legacy Data Center Architecture
Service providers and companies that support data centers are familiar with legacy multi-tiered architectures, as seen in Figure 1.
Figure 1: Legacy Data Center Architecture
The access layer connects servers and other devices to a Layer 2 switch and provides an entry point into the data center. Several access switches are in turn connected to intermediate Layer 2 switches at the aggregation layer (sometimes referred to as the distribution layer) to consolidate traffic. A core layer interconnects the aggregation layer switches. Finally, the core switches are connected to Layer 3 routers in the routing layer to send the aggregated data center traffic to other data centers or a wide area network (WAN), receive external traffic destined for the data center, and interconnect different Layer 2 broadcast domains within the data center.
The problems that exist with the multi-tiered data center architecture include:
- Limited scalability—The demands for electrical power, cooling, cabling, rack space, and port density increase exponentially as the traditional data center expands, which prohibits growth after minimal thresholds are met.
- Inefficient resource usage—Up to 50 percent of switch ports in a legacy data center are used to interconnect different tiers rather than support server and storage connections. In addition, traffic that ideally should move horizontally between servers within a data center often must also be sent vertically up through the tiers to reach a router and down through the tiers to reach the required destination server.
- Increased latency—By requiring the devices at each tier level to perform multiple iterations of packet and frame processing, the data plane traffic takes significantly longer to reach its destination than if the sending and receiving devices were directly connected. This processing overhead results in potentially poor performance for time-sensitive applications, such as voice, video, or financial transactions.
QFX Series QFabric System Architecture
In contrast to legacy multi-tiered data center architectures, the Juniper Networks QFX Series QFabric System architecture provides a simplified networking environment that solves the most challenging issues faced by data center operators. A fabric is a set of devices that act in concert to behave as a single switch. It is a highly scalable, distributed, Layer 2 and Layer 3 networking architecture that provides a high-performance, low-latency, and unified interconnect solution for next-generation data centers as seen in Figure 2.
Figure 2: QFX Series QFabric System Architecture
A QFabric system collapses the traditional multi-tiered data center model into a single tier where all access layer devices (known in the QFabric system model as Node devices) are essentially directly connected to all other access layer devices across a very large scale fabric backplane (known in the QFabric system model as the Interconnect device). Such an architecture enables the consolidation of data center endpoints (such as servers, storage devices, memory, appliances, and routers) and provides better scaling and network virtualization capabilities than traditional data centers.
Essentially, a QFabric system can be viewed as a single, nonblocking, low-latency switch that supports thousands of 10-Gigabit Ethernet ports or 2-Gbps, 4-Gbps, or 8-Gbps Fibre Channel ports to interconnect servers, storage, and the Internet across a high-speed, high-performance fabric. The entire QFabric system is managed as a single entity through a Director group, containing redundant hardware and software components that can be expanded and scaled as the QFabric system grows in size. In addition, the Director group automatically senses when devices are added or removed from the QFabric system and dynamically adjusts the amount of processing resources required to support the system. Such intelligence helps the QFabric system use the minimum amount of power to run the system efficiently, but not waste energy on unused components.
As a result of the QFabric system architecture, data center operators are now realizing the benefits of this next-generation architecture, including:
- Low latency—Because of its inherent advantages in this area, the QFabric system provides an excellent foundation for mission-critical applications such as financial transactions and stock trades, as well as time-sensitive applications such as voice and video.
- Enhanced scalability—The QFabric system can be managed as a single entity and provides support for thousands of data center devices. As Internet traffic continues to grow exponentially with the increase in high-quality video transmissions and rise in the number of mobile devices used worldwide, the QFabric system can keep pace with the demands for bandwidth, applications, and services offered by the data center.
- Virtualization-enabled—The QFabric system was designed to work seamlessly with virtual servers, virtual appliances, and other virtual devices, allowing for even greater scalability, expandability, and rapid deployment of new services than ever before. Migrating to virtual devices also results in significant costs savings, fueled by reduced space requirements, decreased needs for power and cooling, and increased processing capabilities.
- Simplicity—Although the QFabric system can scale to hundreds of devices and thousands of ports, you can still manage the QFabric system as a single system.
- Flexibility—You can deploy the QFabric system as an entire system or in stages.
- Convergence—Because the congestion-free fabric is lossless, all traffic in a QFabric system can be converged onto a single network. As a result, the QFabric system supports Ethernet, Fibre Channel over Ethernet, and native Fibre Channel packets and frames.
Flat, nonblocking, and lossless, the network fabric offered by the QFabric system has the scale and flexibility to meet the needs of small, medium, and large-sized data centers for years to come.
