The Nitty Gritty of the Virtualized Data Center
In the past, there was no real distinction between the switches found in a data center and those used elsewhere in the organization. But this has changed. Today, a cloud data center behaves as a sort of compute fabric, where any server can run any application. As a consequence, the data center infrastructure must adapt to this new, more flexible environment.
Jan 11, 2012 5:00 AM PT
In 2012, we will see the rise of the data center fabric. Within the year, multiple Ethernet fabric technologies will mature into shipping products. This is very timely because server virtualization is now giving rise to a whole new set of data center challenges that traditional switching technologies were not equipped to address. Fabrics will address these challenges with a faster, simpler, more agile infrastructure that will help IT managers virtualize more applications and deliver greater application performance at less cost.
In the past, there was no real distinction between the switches found in a data center and those used elsewhere in the organization. But this has changed. Today, a cloud data center behaves as a sort of compute fabric, where any server can run any application. As a consequence, the data center infrastructure must adapt to this new, more flexible environment. Gartner has characterized this ongoing transition as "an important step in the evolution from box-centric, data center-optimized Ethernet solutions to a fabric-based approach for large data center and cloud infrastructure."
The virtualization problem is this: legacy network switches were designed for the static data center, where servers ran just one application and data flows were well-defined and predictable. In the virtualized data center, however, things are much more dynamic. Servers can run multiple applications, and those applications could run on any server.
Here are several ways that data center fabrics address the virtualization challenge:
- Relieve the east-west traffic problem. Gartner claims that 80 percent of data center network traffic now travels from server to server. Depending on where it's going, that data may travel through multiple layers of infrastructure, creating latency and I/O congestion at choke points along the way. This is referred to as the "east-west" problem.
Data center fabrics will address this with smarter data paths that can traverse the topology much more directly, adding performance and reducing physical complexity.
- Eliminate the spanning tree issue. The Spanning Tree Protocol dates back almost 30 years. Designed to ensure a loop-free topology, it was not intended to build an efficient virtual data center. In fact, it creates major performance challenges by limiting data paths in ways that create congestion.
New solutions employ greater network intelligence to route data in faster, more efficient ways.
- Reduced reliance on VLANs. VLANs, which also date back to the 1980s, segregate network traffic. They also saddle virtualization managers with tedious I/O configuration tasks and present scaling challenges in large data centers.
Data center fabrics will enable simpler, more scalable segregation without reliance on VLANs.
- Connect your VMs in seconds. VMs are quick to deploy but not always easy to connect to other resources. You need a network identity and a data path, both of which will require coordination across multiple teams.
Data center fabrics will let you connect quickly, especially when connecting to other VMs on the same fabric. In that case, traffic can traverse the fabric independent of the general Ethernet production network, thus reducing congestion and eliminating a host of configuration tasks.
- More bandwidth where you really need it: at the server. Traditional networking has skinny pipes down at the server level and progressively fatter pipes as you move up the networking topology. But if 80 percent of the traffic is server-to-server, you actually need more bandwidth down low in the infrastructure.
Putting more bandwidth down at the server level with a data center fabric will help you run more VMs per server.
- FC/Ethernet convergence. Most Fibre Channel shops still use dedicated FC server connections to each server, despite the availability of effective solutions that converge Ethernet and Fibre Channel to a single cable. It has been well-proven that FC and Ethernet can deliver great performance and reliability when combined over a 20 or 40 Gbs link.
In 2012 there will be multiple options to achieve full convergence and save serious money.
- Open solutions for an efficient path forward. Finally, there will be options to achieve all of these goals while fully leveraging the gear you have. With some of the new data center fabric solutions, this transition absolutely can be smooth and incremental, not rip and replace.
In recognition of this growing space, Gartner recently published a new competitive landscape for data center Ethernet switches. Titled "Competitive Landscape: Data Center Ethernet Switches, Worldwide, 2011," the report observed that virtualization and cloud computing have created the need for next-generation fabric-based solutions that provide far more agility and performance.
Gartner stated in the report that "data center switching is a fast-evolving market with a different problem set and several new technologies likely to disrupt legacies."
We are seeing this now as multiple technology providers, both old and new, vie to deliver the next-generation, fabric-based data center architecture.