In Data Centers today there is one constant; perpetual change. That one element is what drives the limits of business and the foundational technologies that enable it. In a bold effort to keep pace with this concept of change, a very flexible and efficient Data Center is needed. A Data Center that can support scalable Layer 2 networks needed for high performance computing (HPC), resource pooling, workload flexibility and virtualization.
The truth is that today our Layer 2 domains are severely hampered by fundamental issues associated with their most basic governing element; spanning tree protocol (STP). In an effort to block links thus avoiding the creation of bridging loops STP is in effect limiting our scalability by creating bandwidth restriction. More often than not even the mechanism used by STP to converge can cause application disruption as STP maneuvers around failed links. This means that the typical approach to address this issue in most networks is to deploy smaller layer 2 domains with Layer 3 boundaries, thus affording STP more stability and scalability. This approach of scaling down the layer 2 domain however will not work in the Data Center because virtualization and HPC applications require larger Layer 2 domains in order to maintain server capacity and high availability in the same VLAN. This quandary was addressed with the creation of Cisco FabricPath.
Designed to address the scaling limitations of Spanning Tree, FabricPath helps create a new generation of scalable and dynamic Data Centers. This NXOS enhancement actually combines the simplicity of Layer 2 switching with the scalability and resilience of Layer 3 routing. FabricPath removes the need for STP and in doing so allows us to create highly scalable Layer 2 ethernet networks that have numerous active and forwarding links. This creates what is known as a FabricPath switching system.
In this new FabricPath switching system traffic will be spread across all available paths, thus significantly increasing available bandwidth. In the event of a link or switch failure, packet forwarding will still continue to take place via any of the remaining links. This multi-path ability will take place automatically, thus ensuring optimum packet delivery even in a failure scenario. Additionally it needs to be noted that FabricPath will work to insure the lowest end-to-end latency by guaranteeing that server-to-server communications occur over the shortest path available in the infrastructure.
Possibly the largest benefit that FabricPath brings us is the elimination of the need for switch ports to be in the same domain. FabricPath allows any port in any VLAN on any switch to exist anywhere in the Data Center, thus increasing flexibility of deployment and in turn simplifying operations. This means that we have here-to-for unheard of scalability in our network. We can now easily add new links to increase bandwidth. We can add more switching fabric it increase switching capacity. FabricPath will automatically recognize these changes. How does FabricPath accomplish this?
Let’s take a closer look.
Keep in mind that FabricPath is a Layer 2 routing technology. As such each FabricPath enabled switch will acquire a unique MAC address which is automatically distributed and maintained within a MAC routing table. Access switches learn the MAC addresses of all the hosts directly attached to it. When host A wants to communicate with host Z, host A sends an Ethernet frame to its attached switch, that switch will then add a FabricPath destination switch header and then routes it across the Layer 2 FabricPath switching system. Once it arrives at the destination switch that switch removes the header and the Ethernet frame is forwarded to host Z.
This means that we can now build highly scalable high bandwidth fabrics. High performance computing applications can take advantage of the inherent low latency, and shortest path non-blocking design needed to support demanding HPC cluster applications.
Even the deployment of applications in this model is greatly simplified. With FabricPath, VLANs can now be extended across the Data Center allowing any application to be supported on any server anywhere in the Data Center. This means that applications are no longer required to be tethered to specific racks. They could be deployed on opposite ends of the Data Center now just as easily as they could on servers in the same rack. This means that a new application could be deployed by simply leveraging any existing available servers or by adding new server capacity anywhere in the Data Center without having to physically move racks or make network changes.
As we can see FabricPath, just one of the innovations NXOS offers, brings us an entire range of opportunities, capabilities and efficiencies that allow us to maximize Data Center flexibility, resiliency and scalability for virtualization and HPC environments. In subsequent articles we will look at other NXOS enhancements.
If you have any questions do not hesitate to ask.
CCIE 35347 (R&S)