Every big network vendor is now touting an architecture that challenges the traditional three-tier design. Cisco and Hewlett-Packard are introducing equipment and design changes that increase the size of the Layer 2 collision domain to enable greater virtual workload mobility. Network managers are being offered a choice between known, proven designs and unknown, flattened architectures. And these new architectures require new switch chassis along with new line and supervisor cards.
No network manager will be fired for adhering to risk-avoiding orthodox networks. The question is, in the face of sweeping changes offered by virtualization and cloud computing, can traditional networks remain competitive?
For example, the widely used and well-understood STP (Spanning Tree Protocol) makes switched networks possible. At a basic level, STP enables switches to be plugged into other switches so that redundant paths are available to prevent a single point of failure, but loops, which are fatal to network operation, are prevented. Creating smaller collision domains inside a larger broadcast domain worked in networks where applications were installed on single, physical host systems. However, when VMware vSphere was introduced, these smaller domains also limited where virtual machines could migrate because VMs must have access to the same subnets on the source and destination physical hosts. Thus, business-continuity strategies are usually limited to a single physical location and are, therefore, much more likely to fail.
Two architectures, one from Cisco and the other from HP, offer a range of hardware, protocol and management changes. It’s worth noting that other network-equipment manufacturers are also offering architecture options including Brocade’s Ethernet fabrics, Extreme’s Open Fabric and Juniper Networks’ just released QFabric.
Starting in 2008, Cisco introduced the Nexus 7000 data center switch chassis, a new operating system called NX-OS and a family of Nexus switches. The FabricPath architecture came the following year and in 2010 OTV (Overlay Transport Virtualization), which extends Layer 2 functionality between data centers, was released. Along with these switch and protocol changes Cisco released its UCS (Unified Computing System) server hardware, thus setting the stage for a substantial change in network architecture and the current competitive struggle with HP.
Last month at Interop, HP announced the A12500, which is roughly analogous to the Cisco 7000. HP also unveiled its FlexNetwork architecture including FlexFabric for the data center. It almost goes without saying that HP has long made data center servers.
Aging protocols are being jostled by emerging network concepts that could morph into the standards that will run on networks in short order. Besides working to move beyond STP, a research project called OpenFlow-an experimental, open standard intended for future switches, routers and wireless access points-seeks to be the basis for software-defined networks where hardware elements support optimized traffic flows that eliminate bottlenecks that current protocols such as OSPF (Open Shortest Path First) can introduce. And all of these protocols will require better network-management oversight.
Cisco is often chided for having too many disparate management tools while HP has a well-established place in the enterprise operations-management field. Both companies are using the network architecture changes as an opportunity to recreate management as a competitive advantage, which is all to the good for IT managers. Cisco is reducing the number of management tools by consolidating functions while HP started shipping IMC5 release (Intelligent Management Center, version 5) at Interop.
The Fabric and the Tree
In the next several years, as early adopters put the new equipment and designs into practice, no word is likely to be more used and abused than “fabric.” Both Cisco and HP use fabric as an integral part of their network designs. In both cases, fabric lacks the concrete, limited meaning found in the storage world. Instead, a network fabric is a densely woven high-speed interconnect between compute and storage in an environment where traffic flows and workloads are dynamically relocated based on priority and available resources.
Cisco and HP both recognize that moving from a tree to a fabric-based architecture will be an evolution. Neither company sees network managers ordering a rip-and-replace strategy. And the new network designs aren’t suited for all business needs. For example, Microsoft Exchange or SharePoint client/server traffic will likely still be best served by current designs. Instead of taking a rip-and-replace approach, both companies have made it clear that they will continue to support three-tier designs that use STP. But IT managers can expect a push from equipment makers to purchase switch line cards that enable the gradual transition to a fabric design. The evolution from tree to fabric favors the edge and core and deemphasizes the aggregation layer.
Cisco and FabricPath
Cisco opened the attack on STP on a variety of fronts when it launched the Nexus product platform. In short order, OTV and FabricPath were released to deal with the Layer 2 domain limits of STP. Around the same time, the VPC (Virtual PortChannel) was introduced on the Nexus 7000, and that enabled all uplink ports to be used to increase bandwidth but without introducing loops.
Cisco introduced FabricPath as a feature in the Nexus 7000 hardware platform. The FabricPath Switching System combines NX-OS software features with the hard capabilities of the Nexus 7000 F1 series module as the basis of its next-generation network architecture.
Hardware modules incorporating FabricPath enable workload mobility such as VMware vMotion in a much larger resource pool, while also gaining significant bandwidth and N+1 redundancy and fast network convergence after a link failure. FabricPath-enabled modules use active-active links between devices-losing the idle ports that spanning tree required-and instead use ECMP (equal-cost multipath) to direct traffic.
First-generation modules can support up to 16-way ECMP, which can be combined with 16-port 10G bps PortChannels, for a total of 2.56T bps between switches. FabricPath builds on technologies that Cisco developed to create OTV and increases performance for enterprises moving to the Nexus 7000 and NX-OS.
HP Releases FlexNetwork
After being a Cisco partner for many years, HP became a fierce rival. Its acquisition of 3Com last year was aimed at shoring up its position against Cisco. HP’s release of the FlexNetwork architecture last month is among its latest competitive salvos.
The data center heart of the architecture is FlexFabric and the HP A10500. In keeping with its legacy, HP has placed unified management at the heart of the FlexNetwork architecture, which is also a distinguishing characteristic of the design. Although the much-worn “single pane of glass” metaphor is central to the IMC5 release, it is useful to know that more than 10 separate modules can be implemented as needed. The network element manager is a basic requirement that IT managers should place at the top of their evaluation list.
The A10500 is one of the main hardware components of the FlexNetwork design but is not expected in the United States until sometime in the third quarter, although the chassis is shipping in China. Thus, I have not yet seen the “proof in the pudding” as it were. Even so, HP appears to have all the basic ingredients to support the FlexNetwork architecture with the hardware and software components that were announced in mid-May.
All told, it’s clear that in the coming decade an evolutionary change in enterprise networks will happen. It’s easy to let the incremental changes slip by. However, it’s unmistakable that virtualization, mobility and cloud computing will soon place stresses on networks that forward-thinking IT managers must consider today.