By Deepak Munjal, HP Networking Strategist
As data centers networks evolve to support virtualization, storage convergence and cloud computing, I know that you are looking for ways not only to improve performance but also minimize complexity. The fact is that over the years, we’ve built data center networks that focused on creating resiliency at the expense of simplicity.
The Spanning Tree Protocol (STP) was created and deployed widely to minimize disruptions in data center networks by offering redundant paths when links or interfaces failed. However, in exchange for this redundancy, we have added quite a bit of management and configuration complexity and many vendors, Cisco specifically, have been complicit in making the problem even worse with proprietary hooks and protocols intended to lock-in customers.
How did we get here?
The Ethernet protocol was designed as a Layer 2 protocol that forced a tree-like topology with no loops. This kept the protocol simple and minimized complexity when locating devices that joined and left the network since there was only one path to any given device. In campus and office environments, this works very well but in the data center where servers are “always-on”, there was a need for a higher level of redundancy when links failed. Thus, Spanning Tree was created as a way to create Layer 2 networks with multiple physical paths while keeping a logical tree topology. When links failed, redundant paths would be activated to continue forwarding packets.
Unfortunately, the process to activate a new link sometimes took seconds if not minutes. While this may have been fine for typical client-server applications, this wasn’t sufficient for latency-sensitive applications like voice or storage which do not tolerate outages this long. Instead of creating a new protocol, vendors have attempted over the past 10 years to minimize this deficiency by adding new hooks and protocols to Spanning Tree. Since there was little standards effort in this space, it allowed networking vendors to create individual solutions to solve this problem that were never intended to be interoperable. Even worse, you saw multiple solutions from the same vendor because different business units within the same company couldn’t agree on a single method either.
Cisco has tried harder to solve this problem than any other vendor. Unfortunately, it has only added complexity, forcing you to adopt solutions that are restricted to specific platforms, each with its own feature set and configuration on a different OS. For example, if you want to run a Catalyst 6500 environment, you would use Virtual Switching System (VSS) to optimize Spanning Tree at the edge of the network. However, if you are using the Nexus platform, you would have to use another protocol call Virtual PortChannel (vPC) to accomplish the same thing. In addition, you need yet another protocol for redundancy in the Layer 3 core called HSRP or VRRP.
Cisco has attempted to address the deficiencies in these protocols with a new protocol called FabricPath but it is only supported on the Nexus 7000 and not the Nexus 5000 or even the recently announced Nexus 3000. This has got to be confusing for you—and explains why many customers are asking for a new approach.
A simpler way—with Intelligent Resilient Framework
HP supports a protocol called Intelligent Resilient Framework (IRF) to solve the problem of creating a resilient, high-performance Layer 2 network and has been available in the industry for almost ten years. The benefits of IRF include the ability to have one protocol for the access and core layers as well as for Layer 2 or Layer 3 networks. IRF actually replaces VSS, vPC, and HSRP/VRRP while scaling beyond just two switches. Failover has been tested at less than 50 milliseconds which is fast enough for almost any application including voice
and storage traffic.
IRF is supported across the entire A-Series platform and can even be used in the campus and across data centers. Configuration is simple and there is no need to learn more than one protocol for each of these environments.
It looks like the industry is finally trying to solve this problem in standards-based way. The IETF if currently working on a protocol called TRILL that will once and for all remove the requirement for Spanning Tree by allowing arbitrary topologies and quick recovery of failed links. HP is committed to supporting TRILL as soon as it is standardized. Until then, IRF can simplify your data center network while maintaining a high level of resiliency.
>> Before replacing existing Cisco switches & routers with more of the same Pause and consider this
>> Calling all triviaholics! Play the #HPFunFacts Twitter Trivia game this week (July 11-14). Details here ~ win $20 gift certificate to Amazon, iTunes or Starbucks.